1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * hosting IBM Z kernel virtual machines (s390x)
4 *
5 * Copyright IBM Corp. 2008, 2020
6 *
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 * Christian Borntraeger <borntraeger@de.ibm.com>
9 * Christian Ehrhardt <ehrhardt@de.ibm.com>
10 * Jason J. Herne <jjherne@us.ibm.com>
11 */
12
13 #define KMSG_COMPONENT "kvm-s390"
14 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
15
16 #include <linux/compiler.h>
17 #include <linux/err.h>
18 #include <linux/fs.h>
19 #include <linux/hrtimer.h>
20 #include <linux/init.h>
21 #include <linux/kvm.h>
22 #include <linux/kvm_host.h>
23 #include <linux/mman.h>
24 #include <linux/module.h>
25 #include <linux/moduleparam.h>
26 #include <linux/random.h>
27 #include <linux/slab.h>
28 #include <linux/timer.h>
29 #include <linux/vmalloc.h>
30 #include <linux/bitmap.h>
31 #include <linux/sched/signal.h>
32 #include <linux/string.h>
33 #include <linux/pgtable.h>
34 #include <linux/mmu_notifier.h>
35
36 #include <asm/access-regs.h>
37 #include <asm/asm-offsets.h>
38 #include <asm/lowcore.h>
39 #include <asm/stp.h>
40 #include <asm/gmap.h>
41 #include <asm/nmi.h>
42 #include <asm/isc.h>
43 #include <asm/sclp.h>
44 #include <asm/cpacf.h>
45 #include <asm/timex.h>
46 #include <asm/fpu.h>
47 #include <asm/ap.h>
48 #include <asm/uv.h>
49 #include "kvm-s390.h"
50 #include "gaccess.h"
51 #include "pci.h"
52
53 #define CREATE_TRACE_POINTS
54 #include "trace.h"
55 #include "trace-s390.h"
56
57 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
58 #define LOCAL_IRQS 32
59 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
60 (KVM_MAX_VCPUS + LOCAL_IRQS))
61
62 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
63 KVM_GENERIC_VM_STATS(),
64 STATS_DESC_COUNTER(VM, inject_io),
65 STATS_DESC_COUNTER(VM, inject_float_mchk),
66 STATS_DESC_COUNTER(VM, inject_pfault_done),
67 STATS_DESC_COUNTER(VM, inject_service_signal),
68 STATS_DESC_COUNTER(VM, inject_virtio),
69 STATS_DESC_COUNTER(VM, aen_forward),
70 STATS_DESC_COUNTER(VM, gmap_shadow_reuse),
71 STATS_DESC_COUNTER(VM, gmap_shadow_create),
72 STATS_DESC_COUNTER(VM, gmap_shadow_r1_entry),
73 STATS_DESC_COUNTER(VM, gmap_shadow_r2_entry),
74 STATS_DESC_COUNTER(VM, gmap_shadow_r3_entry),
75 STATS_DESC_COUNTER(VM, gmap_shadow_sg_entry),
76 STATS_DESC_COUNTER(VM, gmap_shadow_pg_entry),
77 };
78
79 const struct kvm_stats_header kvm_vm_stats_header = {
80 .name_size = KVM_STATS_NAME_SIZE,
81 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
82 .id_offset = sizeof(struct kvm_stats_header),
83 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
84 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
85 sizeof(kvm_vm_stats_desc),
86 };
87
88 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
89 KVM_GENERIC_VCPU_STATS(),
90 STATS_DESC_COUNTER(VCPU, exit_userspace),
91 STATS_DESC_COUNTER(VCPU, exit_null),
92 STATS_DESC_COUNTER(VCPU, exit_external_request),
93 STATS_DESC_COUNTER(VCPU, exit_io_request),
94 STATS_DESC_COUNTER(VCPU, exit_external_interrupt),
95 STATS_DESC_COUNTER(VCPU, exit_stop_request),
96 STATS_DESC_COUNTER(VCPU, exit_validity),
97 STATS_DESC_COUNTER(VCPU, exit_instruction),
98 STATS_DESC_COUNTER(VCPU, exit_pei),
99 STATS_DESC_COUNTER(VCPU, halt_no_poll_steal),
100 STATS_DESC_COUNTER(VCPU, instruction_lctl),
101 STATS_DESC_COUNTER(VCPU, instruction_lctlg),
102 STATS_DESC_COUNTER(VCPU, instruction_stctl),
103 STATS_DESC_COUNTER(VCPU, instruction_stctg),
104 STATS_DESC_COUNTER(VCPU, exit_program_interruption),
105 STATS_DESC_COUNTER(VCPU, exit_instr_and_program),
106 STATS_DESC_COUNTER(VCPU, exit_operation_exception),
107 STATS_DESC_COUNTER(VCPU, deliver_ckc),
108 STATS_DESC_COUNTER(VCPU, deliver_cputm),
109 STATS_DESC_COUNTER(VCPU, deliver_external_call),
110 STATS_DESC_COUNTER(VCPU, deliver_emergency_signal),
111 STATS_DESC_COUNTER(VCPU, deliver_service_signal),
112 STATS_DESC_COUNTER(VCPU, deliver_virtio),
113 STATS_DESC_COUNTER(VCPU, deliver_stop_signal),
114 STATS_DESC_COUNTER(VCPU, deliver_prefix_signal),
115 STATS_DESC_COUNTER(VCPU, deliver_restart_signal),
116 STATS_DESC_COUNTER(VCPU, deliver_program),
117 STATS_DESC_COUNTER(VCPU, deliver_io),
118 STATS_DESC_COUNTER(VCPU, deliver_machine_check),
119 STATS_DESC_COUNTER(VCPU, exit_wait_state),
120 STATS_DESC_COUNTER(VCPU, inject_ckc),
121 STATS_DESC_COUNTER(VCPU, inject_cputm),
122 STATS_DESC_COUNTER(VCPU, inject_external_call),
123 STATS_DESC_COUNTER(VCPU, inject_emergency_signal),
124 STATS_DESC_COUNTER(VCPU, inject_mchk),
125 STATS_DESC_COUNTER(VCPU, inject_pfault_init),
126 STATS_DESC_COUNTER(VCPU, inject_program),
127 STATS_DESC_COUNTER(VCPU, inject_restart),
128 STATS_DESC_COUNTER(VCPU, inject_set_prefix),
129 STATS_DESC_COUNTER(VCPU, inject_stop_signal),
130 STATS_DESC_COUNTER(VCPU, instruction_epsw),
131 STATS_DESC_COUNTER(VCPU, instruction_gs),
132 STATS_DESC_COUNTER(VCPU, instruction_io_other),
133 STATS_DESC_COUNTER(VCPU, instruction_lpsw),
134 STATS_DESC_COUNTER(VCPU, instruction_lpswe),
135 STATS_DESC_COUNTER(VCPU, instruction_lpswey),
136 STATS_DESC_COUNTER(VCPU, instruction_pfmf),
137 STATS_DESC_COUNTER(VCPU, instruction_ptff),
138 STATS_DESC_COUNTER(VCPU, instruction_sck),
139 STATS_DESC_COUNTER(VCPU, instruction_sckpf),
140 STATS_DESC_COUNTER(VCPU, instruction_stidp),
141 STATS_DESC_COUNTER(VCPU, instruction_spx),
142 STATS_DESC_COUNTER(VCPU, instruction_stpx),
143 STATS_DESC_COUNTER(VCPU, instruction_stap),
144 STATS_DESC_COUNTER(VCPU, instruction_iske),
145 STATS_DESC_COUNTER(VCPU, instruction_ri),
146 STATS_DESC_COUNTER(VCPU, instruction_rrbe),
147 STATS_DESC_COUNTER(VCPU, instruction_sske),
148 STATS_DESC_COUNTER(VCPU, instruction_ipte_interlock),
149 STATS_DESC_COUNTER(VCPU, instruction_stsi),
150 STATS_DESC_COUNTER(VCPU, instruction_stfl),
151 STATS_DESC_COUNTER(VCPU, instruction_tb),
152 STATS_DESC_COUNTER(VCPU, instruction_tpi),
153 STATS_DESC_COUNTER(VCPU, instruction_tprot),
154 STATS_DESC_COUNTER(VCPU, instruction_tsch),
155 STATS_DESC_COUNTER(VCPU, instruction_sie),
156 STATS_DESC_COUNTER(VCPU, instruction_essa),
157 STATS_DESC_COUNTER(VCPU, instruction_sthyi),
158 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense),
159 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense_running),
160 STATS_DESC_COUNTER(VCPU, instruction_sigp_external_call),
161 STATS_DESC_COUNTER(VCPU, instruction_sigp_emergency),
162 STATS_DESC_COUNTER(VCPU, instruction_sigp_cond_emergency),
163 STATS_DESC_COUNTER(VCPU, instruction_sigp_start),
164 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop),
165 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop_store_status),
166 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_status),
167 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_adtl_status),
168 STATS_DESC_COUNTER(VCPU, instruction_sigp_arch),
169 STATS_DESC_COUNTER(VCPU, instruction_sigp_prefix),
170 STATS_DESC_COUNTER(VCPU, instruction_sigp_restart),
171 STATS_DESC_COUNTER(VCPU, instruction_sigp_init_cpu_reset),
172 STATS_DESC_COUNTER(VCPU, instruction_sigp_cpu_reset),
173 STATS_DESC_COUNTER(VCPU, instruction_sigp_unknown),
174 STATS_DESC_COUNTER(VCPU, instruction_diagnose_10),
175 STATS_DESC_COUNTER(VCPU, instruction_diagnose_44),
176 STATS_DESC_COUNTER(VCPU, instruction_diagnose_9c),
177 STATS_DESC_COUNTER(VCPU, diag_9c_ignored),
178 STATS_DESC_COUNTER(VCPU, diag_9c_forward),
179 STATS_DESC_COUNTER(VCPU, instruction_diagnose_258),
180 STATS_DESC_COUNTER(VCPU, instruction_diagnose_308),
181 STATS_DESC_COUNTER(VCPU, instruction_diagnose_500),
182 STATS_DESC_COUNTER(VCPU, instruction_diagnose_other),
183 STATS_DESC_COUNTER(VCPU, pfault_sync)
184 };
185
186 const struct kvm_stats_header kvm_vcpu_stats_header = {
187 .name_size = KVM_STATS_NAME_SIZE,
188 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
189 .id_offset = sizeof(struct kvm_stats_header),
190 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
191 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
192 sizeof(kvm_vcpu_stats_desc),
193 };
194
195 /* allow nested virtualization in KVM (if enabled by user space) */
196 static int nested;
197 module_param(nested, int, S_IRUGO);
198 MODULE_PARM_DESC(nested, "Nested virtualization support");
199
200 /* allow 1m huge page guest backing, if !nested */
201 static int hpage;
202 module_param(hpage, int, 0444);
203 MODULE_PARM_DESC(hpage, "1m huge page backing support");
204
205 /* maximum percentage of steal time for polling. >100 is treated like 100 */
206 static u8 halt_poll_max_steal = 10;
207 module_param(halt_poll_max_steal, byte, 0644);
208 MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling");
209
210 /* if set to true, the GISA will be initialized and used if available */
211 static bool use_gisa = true;
212 module_param(use_gisa, bool, 0644);
213 MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it.");
214
215 /* maximum diag9c forwarding per second */
216 unsigned int diag9c_forwarding_hz;
217 module_param(diag9c_forwarding_hz, uint, 0644);
218 MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off");
219
220 /*
221 * allow asynchronous deinit for protected guests; enable by default since
222 * the feature is opt-in anyway
223 */
224 static int async_destroy = 1;
225 module_param(async_destroy, int, 0444);
226 MODULE_PARM_DESC(async_destroy, "Asynchronous destroy for protected guests");
227
228 /*
229 * For now we handle at most 16 double words as this is what the s390 base
230 * kernel handles and stores in the prefix page. If we ever need to go beyond
231 * this, this requires changes to code, but the external uapi can stay.
232 */
233 #define SIZE_INTERNAL 16
234
235 /*
236 * Base feature mask that defines default mask for facilities. Consists of the
237 * defines in FACILITIES_KVM and the non-hypervisor managed bits.
238 */
239 static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
240 /*
241 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
242 * and defines the facilities that can be enabled via a cpu model.
243 */
244 static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
245
kvm_s390_fac_size(void)246 static unsigned long kvm_s390_fac_size(void)
247 {
248 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
249 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
250 BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
251 sizeof(stfle_fac_list));
252
253 return SIZE_INTERNAL;
254 }
255
256 /* available cpu features supported by kvm */
257 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
258 /* available subfunctions indicated via query / "test bit" */
259 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
260
261 static struct gmap_notifier gmap_notifier;
262 static struct gmap_notifier vsie_gmap_notifier;
263 debug_info_t *kvm_s390_dbf;
264 debug_info_t *kvm_s390_dbf_uv;
265
266 /* Section: not file related */
267 /* forward declarations */
268 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
269 unsigned long end);
270 static int sca_switch_to_extended(struct kvm *kvm);
271
kvm_clock_sync_scb(struct kvm_s390_sie_block * scb,u64 delta)272 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
273 {
274 u8 delta_idx = 0;
275
276 /*
277 * The TOD jumps by delta, we have to compensate this by adding
278 * -delta to the epoch.
279 */
280 delta = -delta;
281
282 /* sign-extension - we're adding to signed values below */
283 if ((s64)delta < 0)
284 delta_idx = -1;
285
286 scb->epoch += delta;
287 if (scb->ecd & ECD_MEF) {
288 scb->epdx += delta_idx;
289 if (scb->epoch < delta)
290 scb->epdx += 1;
291 }
292 }
293
294 /*
295 * This callback is executed during stop_machine(). All CPUs are therefore
296 * temporarily stopped. In order not to change guest behavior, we have to
297 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
298 * so a CPU won't be stopped while calculating with the epoch.
299 */
kvm_clock_sync(struct notifier_block * notifier,unsigned long val,void * v)300 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
301 void *v)
302 {
303 struct kvm *kvm;
304 struct kvm_vcpu *vcpu;
305 unsigned long i;
306 unsigned long long *delta = v;
307
308 list_for_each_entry(kvm, &vm_list, vm_list) {
309 kvm_for_each_vcpu(i, vcpu, kvm) {
310 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
311 if (i == 0) {
312 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
313 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
314 }
315 if (vcpu->arch.cputm_enabled)
316 vcpu->arch.cputm_start += *delta;
317 if (vcpu->arch.vsie_block)
318 kvm_clock_sync_scb(vcpu->arch.vsie_block,
319 *delta);
320 }
321 }
322 return NOTIFY_OK;
323 }
324
325 static struct notifier_block kvm_clock_notifier = {
326 .notifier_call = kvm_clock_sync,
327 };
328
allow_cpu_feat(unsigned long nr)329 static void allow_cpu_feat(unsigned long nr)
330 {
331 set_bit_inv(nr, kvm_s390_available_cpu_feat);
332 }
333
plo_test_bit(unsigned char nr)334 static inline int plo_test_bit(unsigned char nr)
335 {
336 unsigned long function = (unsigned long)nr | 0x100;
337 int cc;
338
339 asm volatile(
340 " lgr 0,%[function]\n"
341 /* Parameter registers are ignored for "test bit" */
342 " plo 0,0,0,0(0)\n"
343 " ipm %0\n"
344 " srl %0,28\n"
345 : "=d" (cc)
346 : [function] "d" (function)
347 : "cc", "0");
348 return cc == 0;
349 }
350
__sortl_query(u8 (* query)[32])351 static __always_inline void __sortl_query(u8 (*query)[32])
352 {
353 asm volatile(
354 " lghi 0,0\n"
355 " la 1,%[query]\n"
356 /* Parameter registers are ignored */
357 " .insn rre,0xb9380000,2,4\n"
358 : [query] "=R" (*query)
359 :
360 : "cc", "0", "1");
361 }
362
__dfltcc_query(u8 (* query)[32])363 static __always_inline void __dfltcc_query(u8 (*query)[32])
364 {
365 asm volatile(
366 " lghi 0,0\n"
367 " la 1,%[query]\n"
368 /* Parameter registers are ignored */
369 " .insn rrf,0xb9390000,2,4,6,0\n"
370 : [query] "=R" (*query)
371 :
372 : "cc", "0", "1");
373 }
374
kvm_s390_cpu_feat_init(void)375 static void __init kvm_s390_cpu_feat_init(void)
376 {
377 int i;
378
379 for (i = 0; i < 256; ++i) {
380 if (plo_test_bit(i))
381 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
382 }
383
384 if (test_facility(28)) /* TOD-clock steering */
385 ptff(kvm_s390_available_subfunc.ptff,
386 sizeof(kvm_s390_available_subfunc.ptff),
387 PTFF_QAF);
388
389 if (test_facility(17)) { /* MSA */
390 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
391 kvm_s390_available_subfunc.kmac);
392 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
393 kvm_s390_available_subfunc.kmc);
394 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
395 kvm_s390_available_subfunc.km);
396 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
397 kvm_s390_available_subfunc.kimd);
398 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
399 kvm_s390_available_subfunc.klmd);
400 }
401 if (test_facility(76)) /* MSA3 */
402 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
403 kvm_s390_available_subfunc.pckmo);
404 if (test_facility(77)) { /* MSA4 */
405 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
406 kvm_s390_available_subfunc.kmctr);
407 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
408 kvm_s390_available_subfunc.kmf);
409 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
410 kvm_s390_available_subfunc.kmo);
411 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
412 kvm_s390_available_subfunc.pcc);
413 }
414 if (test_facility(57)) /* MSA5 */
415 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
416 kvm_s390_available_subfunc.ppno);
417
418 if (test_facility(146)) /* MSA8 */
419 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
420 kvm_s390_available_subfunc.kma);
421
422 if (test_facility(155)) /* MSA9 */
423 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *)
424 kvm_s390_available_subfunc.kdsa);
425
426 if (test_facility(150)) /* SORTL */
427 __sortl_query(&kvm_s390_available_subfunc.sortl);
428
429 if (test_facility(151)) /* DFLTCC */
430 __dfltcc_query(&kvm_s390_available_subfunc.dfltcc);
431
432 if (MACHINE_HAS_ESOP)
433 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
434 /*
435 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
436 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
437 */
438 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
439 !test_facility(3) || !nested)
440 return;
441 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
442 if (sclp.has_64bscao)
443 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
444 if (sclp.has_siif)
445 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
446 if (sclp.has_gpere)
447 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
448 if (sclp.has_gsls)
449 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
450 if (sclp.has_ib)
451 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
452 if (sclp.has_cei)
453 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
454 if (sclp.has_ibs)
455 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
456 if (sclp.has_kss)
457 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
458 /*
459 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
460 * all skey handling functions read/set the skey from the PGSTE
461 * instead of the real storage key.
462 *
463 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
464 * pages being detected as preserved although they are resident.
465 *
466 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
467 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
468 *
469 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
470 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
471 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
472 *
473 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
474 * cannot easily shadow the SCA because of the ipte lock.
475 */
476 }
477
__kvm_s390_init(void)478 static int __init __kvm_s390_init(void)
479 {
480 int rc = -ENOMEM;
481
482 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
483 if (!kvm_s390_dbf)
484 return -ENOMEM;
485
486 kvm_s390_dbf_uv = debug_register("kvm-uv", 32, 1, 7 * sizeof(long));
487 if (!kvm_s390_dbf_uv)
488 goto err_kvm_uv;
489
490 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view) ||
491 debug_register_view(kvm_s390_dbf_uv, &debug_sprintf_view))
492 goto err_debug_view;
493
494 kvm_s390_cpu_feat_init();
495
496 /* Register floating interrupt controller interface. */
497 rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
498 if (rc) {
499 pr_err("A FLIC registration call failed with rc=%d\n", rc);
500 goto err_flic;
501 }
502
503 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
504 rc = kvm_s390_pci_init();
505 if (rc) {
506 pr_err("Unable to allocate AIFT for PCI\n");
507 goto err_pci;
508 }
509 }
510
511 rc = kvm_s390_gib_init(GAL_ISC);
512 if (rc)
513 goto err_gib;
514
515 gmap_notifier.notifier_call = kvm_gmap_notifier;
516 gmap_register_pte_notifier(&gmap_notifier);
517 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
518 gmap_register_pte_notifier(&vsie_gmap_notifier);
519 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
520 &kvm_clock_notifier);
521
522 return 0;
523
524 err_gib:
525 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
526 kvm_s390_pci_exit();
527 err_pci:
528 err_flic:
529 err_debug_view:
530 debug_unregister(kvm_s390_dbf_uv);
531 err_kvm_uv:
532 debug_unregister(kvm_s390_dbf);
533 return rc;
534 }
535
__kvm_s390_exit(void)536 static void __kvm_s390_exit(void)
537 {
538 gmap_unregister_pte_notifier(&gmap_notifier);
539 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
540 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
541 &kvm_clock_notifier);
542
543 kvm_s390_gib_destroy();
544 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
545 kvm_s390_pci_exit();
546 debug_unregister(kvm_s390_dbf);
547 debug_unregister(kvm_s390_dbf_uv);
548 }
549
550 /* Section: device related */
kvm_arch_dev_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)551 long kvm_arch_dev_ioctl(struct file *filp,
552 unsigned int ioctl, unsigned long arg)
553 {
554 if (ioctl == KVM_S390_ENABLE_SIE)
555 return s390_enable_sie();
556 return -EINVAL;
557 }
558
kvm_vm_ioctl_check_extension(struct kvm * kvm,long ext)559 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
560 {
561 int r;
562
563 switch (ext) {
564 case KVM_CAP_S390_PSW:
565 case KVM_CAP_S390_GMAP:
566 case KVM_CAP_SYNC_MMU:
567 #ifdef CONFIG_KVM_S390_UCONTROL
568 case KVM_CAP_S390_UCONTROL:
569 #endif
570 case KVM_CAP_ASYNC_PF:
571 case KVM_CAP_SYNC_REGS:
572 case KVM_CAP_ONE_REG:
573 case KVM_CAP_ENABLE_CAP:
574 case KVM_CAP_S390_CSS_SUPPORT:
575 case KVM_CAP_IOEVENTFD:
576 case KVM_CAP_S390_IRQCHIP:
577 case KVM_CAP_VM_ATTRIBUTES:
578 case KVM_CAP_MP_STATE:
579 case KVM_CAP_IMMEDIATE_EXIT:
580 case KVM_CAP_S390_INJECT_IRQ:
581 case KVM_CAP_S390_USER_SIGP:
582 case KVM_CAP_S390_USER_STSI:
583 case KVM_CAP_S390_SKEYS:
584 case KVM_CAP_S390_IRQ_STATE:
585 case KVM_CAP_S390_USER_INSTR0:
586 case KVM_CAP_S390_CMMA_MIGRATION:
587 case KVM_CAP_S390_AIS:
588 case KVM_CAP_S390_AIS_MIGRATION:
589 case KVM_CAP_S390_VCPU_RESETS:
590 case KVM_CAP_SET_GUEST_DEBUG:
591 case KVM_CAP_S390_DIAG318:
592 case KVM_CAP_IRQFD_RESAMPLE:
593 r = 1;
594 break;
595 case KVM_CAP_SET_GUEST_DEBUG2:
596 r = KVM_GUESTDBG_VALID_MASK;
597 break;
598 case KVM_CAP_S390_HPAGE_1M:
599 r = 0;
600 if (hpage && !(kvm && kvm_is_ucontrol(kvm)))
601 r = 1;
602 break;
603 case KVM_CAP_S390_MEM_OP:
604 r = MEM_OP_MAX_SIZE;
605 break;
606 case KVM_CAP_S390_MEM_OP_EXTENSION:
607 /*
608 * Flag bits indicating which extensions are supported.
609 * If r > 0, the base extension must also be supported/indicated,
610 * in order to maintain backwards compatibility.
611 */
612 r = KVM_S390_MEMOP_EXTENSION_CAP_BASE |
613 KVM_S390_MEMOP_EXTENSION_CAP_CMPXCHG;
614 break;
615 case KVM_CAP_NR_VCPUS:
616 case KVM_CAP_MAX_VCPUS:
617 case KVM_CAP_MAX_VCPU_ID:
618 r = KVM_S390_BSCA_CPU_SLOTS;
619 if (!kvm_s390_use_sca_entries())
620 r = KVM_MAX_VCPUS;
621 else if (sclp.has_esca && sclp.has_64bscao)
622 r = KVM_S390_ESCA_CPU_SLOTS;
623 if (ext == KVM_CAP_NR_VCPUS)
624 r = min_t(unsigned int, num_online_cpus(), r);
625 break;
626 case KVM_CAP_S390_COW:
627 r = MACHINE_HAS_ESOP;
628 break;
629 case KVM_CAP_S390_VECTOR_REGISTERS:
630 r = test_facility(129);
631 break;
632 case KVM_CAP_S390_RI:
633 r = test_facility(64);
634 break;
635 case KVM_CAP_S390_GS:
636 r = test_facility(133);
637 break;
638 case KVM_CAP_S390_BPB:
639 r = test_facility(82);
640 break;
641 case KVM_CAP_S390_PROTECTED_ASYNC_DISABLE:
642 r = async_destroy && is_prot_virt_host();
643 break;
644 case KVM_CAP_S390_PROTECTED:
645 r = is_prot_virt_host();
646 break;
647 case KVM_CAP_S390_PROTECTED_DUMP: {
648 u64 pv_cmds_dump[] = {
649 BIT_UVC_CMD_DUMP_INIT,
650 BIT_UVC_CMD_DUMP_CONFIG_STOR_STATE,
651 BIT_UVC_CMD_DUMP_CPU,
652 BIT_UVC_CMD_DUMP_COMPLETE,
653 };
654 int i;
655
656 r = is_prot_virt_host();
657
658 for (i = 0; i < ARRAY_SIZE(pv_cmds_dump); i++) {
659 if (!test_bit_inv(pv_cmds_dump[i],
660 (unsigned long *)&uv_info.inst_calls_list)) {
661 r = 0;
662 break;
663 }
664 }
665 break;
666 }
667 case KVM_CAP_S390_ZPCI_OP:
668 r = kvm_s390_pci_interp_allowed();
669 break;
670 case KVM_CAP_S390_CPU_TOPOLOGY:
671 r = test_facility(11);
672 break;
673 default:
674 r = 0;
675 }
676 return r;
677 }
678
kvm_arch_sync_dirty_log(struct kvm * kvm,struct kvm_memory_slot * memslot)679 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
680 {
681 int i;
682 gfn_t cur_gfn, last_gfn;
683 unsigned long gaddr, vmaddr;
684 struct gmap *gmap = kvm->arch.gmap;
685 DECLARE_BITMAP(bitmap, _PAGE_ENTRIES);
686
687 /* Loop over all guest segments */
688 cur_gfn = memslot->base_gfn;
689 last_gfn = memslot->base_gfn + memslot->npages;
690 for (; cur_gfn <= last_gfn; cur_gfn += _PAGE_ENTRIES) {
691 gaddr = gfn_to_gpa(cur_gfn);
692 vmaddr = gfn_to_hva_memslot(memslot, cur_gfn);
693 if (kvm_is_error_hva(vmaddr))
694 continue;
695
696 bitmap_zero(bitmap, _PAGE_ENTRIES);
697 gmap_sync_dirty_log_pmd(gmap, bitmap, gaddr, vmaddr);
698 for (i = 0; i < _PAGE_ENTRIES; i++) {
699 if (test_bit(i, bitmap))
700 mark_page_dirty(kvm, cur_gfn + i);
701 }
702
703 if (fatal_signal_pending(current))
704 return;
705 cond_resched();
706 }
707 }
708
709 /* Section: vm related */
710 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
711
712 /*
713 * Get (and clear) the dirty memory log for a memory slot.
714 */
kvm_vm_ioctl_get_dirty_log(struct kvm * kvm,struct kvm_dirty_log * log)715 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
716 struct kvm_dirty_log *log)
717 {
718 int r;
719 unsigned long n;
720 struct kvm_memory_slot *memslot;
721 int is_dirty;
722
723 if (kvm_is_ucontrol(kvm))
724 return -EINVAL;
725
726 mutex_lock(&kvm->slots_lock);
727
728 r = -EINVAL;
729 if (log->slot >= KVM_USER_MEM_SLOTS)
730 goto out;
731
732 r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
733 if (r)
734 goto out;
735
736 /* Clear the dirty log */
737 if (is_dirty) {
738 n = kvm_dirty_bitmap_bytes(memslot);
739 memset(memslot->dirty_bitmap, 0, n);
740 }
741 r = 0;
742 out:
743 mutex_unlock(&kvm->slots_lock);
744 return r;
745 }
746
icpt_operexc_on_all_vcpus(struct kvm * kvm)747 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
748 {
749 unsigned long i;
750 struct kvm_vcpu *vcpu;
751
752 kvm_for_each_vcpu(i, vcpu, kvm) {
753 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
754 }
755 }
756
kvm_vm_ioctl_enable_cap(struct kvm * kvm,struct kvm_enable_cap * cap)757 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
758 {
759 int r;
760
761 if (cap->flags)
762 return -EINVAL;
763
764 switch (cap->cap) {
765 case KVM_CAP_S390_IRQCHIP:
766 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
767 kvm->arch.use_irqchip = 1;
768 r = 0;
769 break;
770 case KVM_CAP_S390_USER_SIGP:
771 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
772 kvm->arch.user_sigp = 1;
773 r = 0;
774 break;
775 case KVM_CAP_S390_VECTOR_REGISTERS:
776 mutex_lock(&kvm->lock);
777 if (kvm->created_vcpus) {
778 r = -EBUSY;
779 } else if (cpu_has_vx()) {
780 set_kvm_facility(kvm->arch.model.fac_mask, 129);
781 set_kvm_facility(kvm->arch.model.fac_list, 129);
782 if (test_facility(134)) {
783 set_kvm_facility(kvm->arch.model.fac_mask, 134);
784 set_kvm_facility(kvm->arch.model.fac_list, 134);
785 }
786 if (test_facility(135)) {
787 set_kvm_facility(kvm->arch.model.fac_mask, 135);
788 set_kvm_facility(kvm->arch.model.fac_list, 135);
789 }
790 if (test_facility(148)) {
791 set_kvm_facility(kvm->arch.model.fac_mask, 148);
792 set_kvm_facility(kvm->arch.model.fac_list, 148);
793 }
794 if (test_facility(152)) {
795 set_kvm_facility(kvm->arch.model.fac_mask, 152);
796 set_kvm_facility(kvm->arch.model.fac_list, 152);
797 }
798 if (test_facility(192)) {
799 set_kvm_facility(kvm->arch.model.fac_mask, 192);
800 set_kvm_facility(kvm->arch.model.fac_list, 192);
801 }
802 r = 0;
803 } else
804 r = -EINVAL;
805 mutex_unlock(&kvm->lock);
806 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
807 r ? "(not available)" : "(success)");
808 break;
809 case KVM_CAP_S390_RI:
810 r = -EINVAL;
811 mutex_lock(&kvm->lock);
812 if (kvm->created_vcpus) {
813 r = -EBUSY;
814 } else if (test_facility(64)) {
815 set_kvm_facility(kvm->arch.model.fac_mask, 64);
816 set_kvm_facility(kvm->arch.model.fac_list, 64);
817 r = 0;
818 }
819 mutex_unlock(&kvm->lock);
820 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
821 r ? "(not available)" : "(success)");
822 break;
823 case KVM_CAP_S390_AIS:
824 mutex_lock(&kvm->lock);
825 if (kvm->created_vcpus) {
826 r = -EBUSY;
827 } else {
828 set_kvm_facility(kvm->arch.model.fac_mask, 72);
829 set_kvm_facility(kvm->arch.model.fac_list, 72);
830 r = 0;
831 }
832 mutex_unlock(&kvm->lock);
833 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
834 r ? "(not available)" : "(success)");
835 break;
836 case KVM_CAP_S390_GS:
837 r = -EINVAL;
838 mutex_lock(&kvm->lock);
839 if (kvm->created_vcpus) {
840 r = -EBUSY;
841 } else if (test_facility(133)) {
842 set_kvm_facility(kvm->arch.model.fac_mask, 133);
843 set_kvm_facility(kvm->arch.model.fac_list, 133);
844 r = 0;
845 }
846 mutex_unlock(&kvm->lock);
847 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
848 r ? "(not available)" : "(success)");
849 break;
850 case KVM_CAP_S390_HPAGE_1M:
851 mutex_lock(&kvm->lock);
852 if (kvm->created_vcpus)
853 r = -EBUSY;
854 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm))
855 r = -EINVAL;
856 else {
857 r = 0;
858 mmap_write_lock(kvm->mm);
859 kvm->mm->context.allow_gmap_hpage_1m = 1;
860 mmap_write_unlock(kvm->mm);
861 /*
862 * We might have to create fake 4k page
863 * tables. To avoid that the hardware works on
864 * stale PGSTEs, we emulate these instructions.
865 */
866 kvm->arch.use_skf = 0;
867 kvm->arch.use_pfmfi = 0;
868 }
869 mutex_unlock(&kvm->lock);
870 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s",
871 r ? "(not available)" : "(success)");
872 break;
873 case KVM_CAP_S390_USER_STSI:
874 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
875 kvm->arch.user_stsi = 1;
876 r = 0;
877 break;
878 case KVM_CAP_S390_USER_INSTR0:
879 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
880 kvm->arch.user_instr0 = 1;
881 icpt_operexc_on_all_vcpus(kvm);
882 r = 0;
883 break;
884 case KVM_CAP_S390_CPU_TOPOLOGY:
885 r = -EINVAL;
886 mutex_lock(&kvm->lock);
887 if (kvm->created_vcpus) {
888 r = -EBUSY;
889 } else if (test_facility(11)) {
890 set_kvm_facility(kvm->arch.model.fac_mask, 11);
891 set_kvm_facility(kvm->arch.model.fac_list, 11);
892 r = 0;
893 }
894 mutex_unlock(&kvm->lock);
895 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_CPU_TOPOLOGY %s",
896 r ? "(not available)" : "(success)");
897 break;
898 default:
899 r = -EINVAL;
900 break;
901 }
902 return r;
903 }
904
kvm_s390_get_mem_control(struct kvm * kvm,struct kvm_device_attr * attr)905 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
906 {
907 int ret;
908
909 switch (attr->attr) {
910 case KVM_S390_VM_MEM_LIMIT_SIZE:
911 ret = 0;
912 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
913 kvm->arch.mem_limit);
914 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
915 ret = -EFAULT;
916 break;
917 default:
918 ret = -ENXIO;
919 break;
920 }
921 return ret;
922 }
923
kvm_s390_set_mem_control(struct kvm * kvm,struct kvm_device_attr * attr)924 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
925 {
926 int ret;
927 unsigned int idx;
928 switch (attr->attr) {
929 case KVM_S390_VM_MEM_ENABLE_CMMA:
930 ret = -ENXIO;
931 if (!sclp.has_cmma)
932 break;
933
934 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
935 mutex_lock(&kvm->lock);
936 if (kvm->created_vcpus)
937 ret = -EBUSY;
938 else if (kvm->mm->context.allow_gmap_hpage_1m)
939 ret = -EINVAL;
940 else {
941 kvm->arch.use_cmma = 1;
942 /* Not compatible with cmma. */
943 kvm->arch.use_pfmfi = 0;
944 ret = 0;
945 }
946 mutex_unlock(&kvm->lock);
947 break;
948 case KVM_S390_VM_MEM_CLR_CMMA:
949 ret = -ENXIO;
950 if (!sclp.has_cmma)
951 break;
952 ret = -EINVAL;
953 if (!kvm->arch.use_cmma)
954 break;
955
956 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
957 mutex_lock(&kvm->lock);
958 idx = srcu_read_lock(&kvm->srcu);
959 s390_reset_cmma(kvm->arch.gmap->mm);
960 srcu_read_unlock(&kvm->srcu, idx);
961 mutex_unlock(&kvm->lock);
962 ret = 0;
963 break;
964 case KVM_S390_VM_MEM_LIMIT_SIZE: {
965 unsigned long new_limit;
966
967 if (kvm_is_ucontrol(kvm))
968 return -EINVAL;
969
970 if (get_user(new_limit, (u64 __user *)attr->addr))
971 return -EFAULT;
972
973 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
974 new_limit > kvm->arch.mem_limit)
975 return -E2BIG;
976
977 if (!new_limit)
978 return -EINVAL;
979
980 /* gmap_create takes last usable address */
981 if (new_limit != KVM_S390_NO_MEM_LIMIT)
982 new_limit -= 1;
983
984 ret = -EBUSY;
985 mutex_lock(&kvm->lock);
986 if (!kvm->created_vcpus) {
987 /* gmap_create will round the limit up */
988 struct gmap *new = gmap_create(current->mm, new_limit);
989
990 if (!new) {
991 ret = -ENOMEM;
992 } else {
993 gmap_remove(kvm->arch.gmap);
994 new->private = kvm;
995 kvm->arch.gmap = new;
996 ret = 0;
997 }
998 }
999 mutex_unlock(&kvm->lock);
1000 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
1001 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
1002 (void *) kvm->arch.gmap->asce);
1003 break;
1004 }
1005 default:
1006 ret = -ENXIO;
1007 break;
1008 }
1009 return ret;
1010 }
1011
1012 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
1013
kvm_s390_vcpu_crypto_reset_all(struct kvm * kvm)1014 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm)
1015 {
1016 struct kvm_vcpu *vcpu;
1017 unsigned long i;
1018
1019 kvm_s390_vcpu_block_all(kvm);
1020
1021 kvm_for_each_vcpu(i, vcpu, kvm) {
1022 kvm_s390_vcpu_crypto_setup(vcpu);
1023 /* recreate the shadow crycb by leaving the VSIE handler */
1024 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1025 }
1026
1027 kvm_s390_vcpu_unblock_all(kvm);
1028 }
1029
kvm_s390_vm_set_crypto(struct kvm * kvm,struct kvm_device_attr * attr)1030 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
1031 {
1032 mutex_lock(&kvm->lock);
1033 switch (attr->attr) {
1034 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1035 if (!test_kvm_facility(kvm, 76)) {
1036 mutex_unlock(&kvm->lock);
1037 return -EINVAL;
1038 }
1039 get_random_bytes(
1040 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1041 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1042 kvm->arch.crypto.aes_kw = 1;
1043 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
1044 break;
1045 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1046 if (!test_kvm_facility(kvm, 76)) {
1047 mutex_unlock(&kvm->lock);
1048 return -EINVAL;
1049 }
1050 get_random_bytes(
1051 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1052 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1053 kvm->arch.crypto.dea_kw = 1;
1054 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
1055 break;
1056 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1057 if (!test_kvm_facility(kvm, 76)) {
1058 mutex_unlock(&kvm->lock);
1059 return -EINVAL;
1060 }
1061 kvm->arch.crypto.aes_kw = 0;
1062 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
1063 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1064 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
1065 break;
1066 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1067 if (!test_kvm_facility(kvm, 76)) {
1068 mutex_unlock(&kvm->lock);
1069 return -EINVAL;
1070 }
1071 kvm->arch.crypto.dea_kw = 0;
1072 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
1073 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1074 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
1075 break;
1076 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
1077 if (!ap_instructions_available()) {
1078 mutex_unlock(&kvm->lock);
1079 return -EOPNOTSUPP;
1080 }
1081 kvm->arch.crypto.apie = 1;
1082 break;
1083 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
1084 if (!ap_instructions_available()) {
1085 mutex_unlock(&kvm->lock);
1086 return -EOPNOTSUPP;
1087 }
1088 kvm->arch.crypto.apie = 0;
1089 break;
1090 default:
1091 mutex_unlock(&kvm->lock);
1092 return -ENXIO;
1093 }
1094
1095 kvm_s390_vcpu_crypto_reset_all(kvm);
1096 mutex_unlock(&kvm->lock);
1097 return 0;
1098 }
1099
kvm_s390_vcpu_pci_setup(struct kvm_vcpu * vcpu)1100 static void kvm_s390_vcpu_pci_setup(struct kvm_vcpu *vcpu)
1101 {
1102 /* Only set the ECB bits after guest requests zPCI interpretation */
1103 if (!vcpu->kvm->arch.use_zpci_interp)
1104 return;
1105
1106 vcpu->arch.sie_block->ecb2 |= ECB2_ZPCI_LSI;
1107 vcpu->arch.sie_block->ecb3 |= ECB3_AISII + ECB3_AISI;
1108 }
1109
kvm_s390_vcpu_pci_enable_interp(struct kvm * kvm)1110 void kvm_s390_vcpu_pci_enable_interp(struct kvm *kvm)
1111 {
1112 struct kvm_vcpu *vcpu;
1113 unsigned long i;
1114
1115 lockdep_assert_held(&kvm->lock);
1116
1117 if (!kvm_s390_pci_interp_allowed())
1118 return;
1119
1120 /*
1121 * If host is configured for PCI and the necessary facilities are
1122 * available, turn on interpretation for the life of this guest
1123 */
1124 kvm->arch.use_zpci_interp = 1;
1125
1126 kvm_s390_vcpu_block_all(kvm);
1127
1128 kvm_for_each_vcpu(i, vcpu, kvm) {
1129 kvm_s390_vcpu_pci_setup(vcpu);
1130 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1131 }
1132
1133 kvm_s390_vcpu_unblock_all(kvm);
1134 }
1135
kvm_s390_sync_request_broadcast(struct kvm * kvm,int req)1136 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
1137 {
1138 unsigned long cx;
1139 struct kvm_vcpu *vcpu;
1140
1141 kvm_for_each_vcpu(cx, vcpu, kvm)
1142 kvm_s390_sync_request(req, vcpu);
1143 }
1144
1145 /*
1146 * Must be called with kvm->srcu held to avoid races on memslots, and with
1147 * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
1148 */
kvm_s390_vm_start_migration(struct kvm * kvm)1149 static int kvm_s390_vm_start_migration(struct kvm *kvm)
1150 {
1151 struct kvm_memory_slot *ms;
1152 struct kvm_memslots *slots;
1153 unsigned long ram_pages = 0;
1154 int bkt;
1155
1156 /* migration mode already enabled */
1157 if (kvm->arch.migration_mode)
1158 return 0;
1159 slots = kvm_memslots(kvm);
1160 if (!slots || kvm_memslots_empty(slots))
1161 return -EINVAL;
1162
1163 if (!kvm->arch.use_cmma) {
1164 kvm->arch.migration_mode = 1;
1165 return 0;
1166 }
1167 /* mark all the pages in active slots as dirty */
1168 kvm_for_each_memslot(ms, bkt, slots) {
1169 if (!ms->dirty_bitmap)
1170 return -EINVAL;
1171 /*
1172 * The second half of the bitmap is only used on x86,
1173 * and would be wasted otherwise, so we put it to good
1174 * use here to keep track of the state of the storage
1175 * attributes.
1176 */
1177 memset(kvm_second_dirty_bitmap(ms), 0xff, kvm_dirty_bitmap_bytes(ms));
1178 ram_pages += ms->npages;
1179 }
1180 atomic64_set(&kvm->arch.cmma_dirty_pages, ram_pages);
1181 kvm->arch.migration_mode = 1;
1182 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
1183 return 0;
1184 }
1185
1186 /*
1187 * Must be called with kvm->slots_lock to avoid races with ourselves and
1188 * kvm_s390_vm_start_migration.
1189 */
kvm_s390_vm_stop_migration(struct kvm * kvm)1190 static int kvm_s390_vm_stop_migration(struct kvm *kvm)
1191 {
1192 /* migration mode already disabled */
1193 if (!kvm->arch.migration_mode)
1194 return 0;
1195 kvm->arch.migration_mode = 0;
1196 if (kvm->arch.use_cmma)
1197 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
1198 return 0;
1199 }
1200
kvm_s390_vm_set_migration(struct kvm * kvm,struct kvm_device_attr * attr)1201 static int kvm_s390_vm_set_migration(struct kvm *kvm,
1202 struct kvm_device_attr *attr)
1203 {
1204 int res = -ENXIO;
1205
1206 mutex_lock(&kvm->slots_lock);
1207 switch (attr->attr) {
1208 case KVM_S390_VM_MIGRATION_START:
1209 res = kvm_s390_vm_start_migration(kvm);
1210 break;
1211 case KVM_S390_VM_MIGRATION_STOP:
1212 res = kvm_s390_vm_stop_migration(kvm);
1213 break;
1214 default:
1215 break;
1216 }
1217 mutex_unlock(&kvm->slots_lock);
1218
1219 return res;
1220 }
1221
kvm_s390_vm_get_migration(struct kvm * kvm,struct kvm_device_attr * attr)1222 static int kvm_s390_vm_get_migration(struct kvm *kvm,
1223 struct kvm_device_attr *attr)
1224 {
1225 u64 mig = kvm->arch.migration_mode;
1226
1227 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
1228 return -ENXIO;
1229
1230 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
1231 return -EFAULT;
1232 return 0;
1233 }
1234
1235 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod);
1236
kvm_s390_set_tod_ext(struct kvm * kvm,struct kvm_device_attr * attr)1237 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1238 {
1239 struct kvm_s390_vm_tod_clock gtod;
1240
1241 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
1242 return -EFAULT;
1243
1244 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
1245 return -EINVAL;
1246 __kvm_s390_set_tod_clock(kvm, >od);
1247
1248 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
1249 gtod.epoch_idx, gtod.tod);
1250
1251 return 0;
1252 }
1253
kvm_s390_set_tod_high(struct kvm * kvm,struct kvm_device_attr * attr)1254 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1255 {
1256 u8 gtod_high;
1257
1258 if (copy_from_user(>od_high, (void __user *)attr->addr,
1259 sizeof(gtod_high)))
1260 return -EFAULT;
1261
1262 if (gtod_high != 0)
1263 return -EINVAL;
1264 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
1265
1266 return 0;
1267 }
1268
kvm_s390_set_tod_low(struct kvm * kvm,struct kvm_device_attr * attr)1269 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1270 {
1271 struct kvm_s390_vm_tod_clock gtod = { 0 };
1272
1273 if (copy_from_user(>od.tod, (void __user *)attr->addr,
1274 sizeof(gtod.tod)))
1275 return -EFAULT;
1276
1277 __kvm_s390_set_tod_clock(kvm, >od);
1278 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1279 return 0;
1280 }
1281
kvm_s390_set_tod(struct kvm * kvm,struct kvm_device_attr * attr)1282 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1283 {
1284 int ret;
1285
1286 if (attr->flags)
1287 return -EINVAL;
1288
1289 mutex_lock(&kvm->lock);
1290 /*
1291 * For protected guests, the TOD is managed by the ultravisor, so trying
1292 * to change it will never bring the expected results.
1293 */
1294 if (kvm_s390_pv_is_protected(kvm)) {
1295 ret = -EOPNOTSUPP;
1296 goto out_unlock;
1297 }
1298
1299 switch (attr->attr) {
1300 case KVM_S390_VM_TOD_EXT:
1301 ret = kvm_s390_set_tod_ext(kvm, attr);
1302 break;
1303 case KVM_S390_VM_TOD_HIGH:
1304 ret = kvm_s390_set_tod_high(kvm, attr);
1305 break;
1306 case KVM_S390_VM_TOD_LOW:
1307 ret = kvm_s390_set_tod_low(kvm, attr);
1308 break;
1309 default:
1310 ret = -ENXIO;
1311 break;
1312 }
1313
1314 out_unlock:
1315 mutex_unlock(&kvm->lock);
1316 return ret;
1317 }
1318
kvm_s390_get_tod_clock(struct kvm * kvm,struct kvm_s390_vm_tod_clock * gtod)1319 static void kvm_s390_get_tod_clock(struct kvm *kvm,
1320 struct kvm_s390_vm_tod_clock *gtod)
1321 {
1322 union tod_clock clk;
1323
1324 preempt_disable();
1325
1326 store_tod_clock_ext(&clk);
1327
1328 gtod->tod = clk.tod + kvm->arch.epoch;
1329 gtod->epoch_idx = 0;
1330 if (test_kvm_facility(kvm, 139)) {
1331 gtod->epoch_idx = clk.ei + kvm->arch.epdx;
1332 if (gtod->tod < clk.tod)
1333 gtod->epoch_idx += 1;
1334 }
1335
1336 preempt_enable();
1337 }
1338
kvm_s390_get_tod_ext(struct kvm * kvm,struct kvm_device_attr * attr)1339 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1340 {
1341 struct kvm_s390_vm_tod_clock gtod;
1342
1343 memset(>od, 0, sizeof(gtod));
1344 kvm_s390_get_tod_clock(kvm, >od);
1345 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1346 return -EFAULT;
1347
1348 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1349 gtod.epoch_idx, gtod.tod);
1350 return 0;
1351 }
1352
kvm_s390_get_tod_high(struct kvm * kvm,struct kvm_device_attr * attr)1353 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1354 {
1355 u8 gtod_high = 0;
1356
1357 if (copy_to_user((void __user *)attr->addr, >od_high,
1358 sizeof(gtod_high)))
1359 return -EFAULT;
1360 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1361
1362 return 0;
1363 }
1364
kvm_s390_get_tod_low(struct kvm * kvm,struct kvm_device_attr * attr)1365 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1366 {
1367 u64 gtod;
1368
1369 gtod = kvm_s390_get_tod_clock_fast(kvm);
1370 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1371 return -EFAULT;
1372 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1373
1374 return 0;
1375 }
1376
kvm_s390_get_tod(struct kvm * kvm,struct kvm_device_attr * attr)1377 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1378 {
1379 int ret;
1380
1381 if (attr->flags)
1382 return -EINVAL;
1383
1384 switch (attr->attr) {
1385 case KVM_S390_VM_TOD_EXT:
1386 ret = kvm_s390_get_tod_ext(kvm, attr);
1387 break;
1388 case KVM_S390_VM_TOD_HIGH:
1389 ret = kvm_s390_get_tod_high(kvm, attr);
1390 break;
1391 case KVM_S390_VM_TOD_LOW:
1392 ret = kvm_s390_get_tod_low(kvm, attr);
1393 break;
1394 default:
1395 ret = -ENXIO;
1396 break;
1397 }
1398 return ret;
1399 }
1400
kvm_s390_set_processor(struct kvm * kvm,struct kvm_device_attr * attr)1401 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1402 {
1403 struct kvm_s390_vm_cpu_processor *proc;
1404 u16 lowest_ibc, unblocked_ibc;
1405 int ret = 0;
1406
1407 mutex_lock(&kvm->lock);
1408 if (kvm->created_vcpus) {
1409 ret = -EBUSY;
1410 goto out;
1411 }
1412 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1413 if (!proc) {
1414 ret = -ENOMEM;
1415 goto out;
1416 }
1417 if (!copy_from_user(proc, (void __user *)attr->addr,
1418 sizeof(*proc))) {
1419 kvm->arch.model.cpuid = proc->cpuid;
1420 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1421 unblocked_ibc = sclp.ibc & 0xfff;
1422 if (lowest_ibc && proc->ibc) {
1423 if (proc->ibc > unblocked_ibc)
1424 kvm->arch.model.ibc = unblocked_ibc;
1425 else if (proc->ibc < lowest_ibc)
1426 kvm->arch.model.ibc = lowest_ibc;
1427 else
1428 kvm->arch.model.ibc = proc->ibc;
1429 }
1430 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1431 S390_ARCH_FAC_LIST_SIZE_BYTE);
1432 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1433 kvm->arch.model.ibc,
1434 kvm->arch.model.cpuid);
1435 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1436 kvm->arch.model.fac_list[0],
1437 kvm->arch.model.fac_list[1],
1438 kvm->arch.model.fac_list[2]);
1439 } else
1440 ret = -EFAULT;
1441 kfree(proc);
1442 out:
1443 mutex_unlock(&kvm->lock);
1444 return ret;
1445 }
1446
kvm_s390_set_processor_feat(struct kvm * kvm,struct kvm_device_attr * attr)1447 static int kvm_s390_set_processor_feat(struct kvm *kvm,
1448 struct kvm_device_attr *attr)
1449 {
1450 struct kvm_s390_vm_cpu_feat data;
1451
1452 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1453 return -EFAULT;
1454 if (!bitmap_subset((unsigned long *) data.feat,
1455 kvm_s390_available_cpu_feat,
1456 KVM_S390_VM_CPU_FEAT_NR_BITS))
1457 return -EINVAL;
1458
1459 mutex_lock(&kvm->lock);
1460 if (kvm->created_vcpus) {
1461 mutex_unlock(&kvm->lock);
1462 return -EBUSY;
1463 }
1464 bitmap_from_arr64(kvm->arch.cpu_feat, data.feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1465 mutex_unlock(&kvm->lock);
1466 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1467 data.feat[0],
1468 data.feat[1],
1469 data.feat[2]);
1470 return 0;
1471 }
1472
kvm_s390_set_processor_subfunc(struct kvm * kvm,struct kvm_device_attr * attr)1473 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1474 struct kvm_device_attr *attr)
1475 {
1476 mutex_lock(&kvm->lock);
1477 if (kvm->created_vcpus) {
1478 mutex_unlock(&kvm->lock);
1479 return -EBUSY;
1480 }
1481
1482 if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr,
1483 sizeof(struct kvm_s390_vm_cpu_subfunc))) {
1484 mutex_unlock(&kvm->lock);
1485 return -EFAULT;
1486 }
1487 mutex_unlock(&kvm->lock);
1488
1489 VM_EVENT(kvm, 3, "SET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1490 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1491 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1492 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1493 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1494 VM_EVENT(kvm, 3, "SET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1495 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1496 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1497 VM_EVENT(kvm, 3, "SET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1498 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1499 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1500 VM_EVENT(kvm, 3, "SET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1501 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1502 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1503 VM_EVENT(kvm, 3, "SET: guest KM subfunc 0x%16.16lx.%16.16lx",
1504 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1505 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1506 VM_EVENT(kvm, 3, "SET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1507 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1508 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1509 VM_EVENT(kvm, 3, "SET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1510 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1511 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1512 VM_EVENT(kvm, 3, "SET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1513 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1514 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1515 VM_EVENT(kvm, 3, "SET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1516 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1517 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1518 VM_EVENT(kvm, 3, "SET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1519 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1520 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1521 VM_EVENT(kvm, 3, "SET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1522 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1523 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1524 VM_EVENT(kvm, 3, "SET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1525 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1526 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1527 VM_EVENT(kvm, 3, "SET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1528 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1529 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1530 VM_EVENT(kvm, 3, "SET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1531 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1532 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1533 VM_EVENT(kvm, 3, "SET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1534 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1535 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1536 VM_EVENT(kvm, 3, "SET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1537 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1538 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1539 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1540 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1541 VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1542 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1543 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1544 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1545 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1546
1547 return 0;
1548 }
1549
1550 #define KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK \
1551 ( \
1552 ((struct kvm_s390_vm_cpu_uv_feat){ \
1553 .ap = 1, \
1554 .ap_intr = 1, \
1555 }) \
1556 .feat \
1557 )
1558
kvm_s390_set_uv_feat(struct kvm * kvm,struct kvm_device_attr * attr)1559 static int kvm_s390_set_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1560 {
1561 struct kvm_s390_vm_cpu_uv_feat __user *ptr = (void __user *)attr->addr;
1562 unsigned long data, filter;
1563
1564 filter = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1565 if (get_user(data, &ptr->feat))
1566 return -EFAULT;
1567 if (!bitmap_subset(&data, &filter, KVM_S390_VM_CPU_UV_FEAT_NR_BITS))
1568 return -EINVAL;
1569
1570 mutex_lock(&kvm->lock);
1571 if (kvm->created_vcpus) {
1572 mutex_unlock(&kvm->lock);
1573 return -EBUSY;
1574 }
1575 kvm->arch.model.uv_feat_guest.feat = data;
1576 mutex_unlock(&kvm->lock);
1577
1578 VM_EVENT(kvm, 3, "SET: guest UV-feat: 0x%16.16lx", data);
1579
1580 return 0;
1581 }
1582
kvm_s390_set_cpu_model(struct kvm * kvm,struct kvm_device_attr * attr)1583 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1584 {
1585 int ret = -ENXIO;
1586
1587 switch (attr->attr) {
1588 case KVM_S390_VM_CPU_PROCESSOR:
1589 ret = kvm_s390_set_processor(kvm, attr);
1590 break;
1591 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1592 ret = kvm_s390_set_processor_feat(kvm, attr);
1593 break;
1594 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1595 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1596 break;
1597 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1598 ret = kvm_s390_set_uv_feat(kvm, attr);
1599 break;
1600 }
1601 return ret;
1602 }
1603
kvm_s390_get_processor(struct kvm * kvm,struct kvm_device_attr * attr)1604 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1605 {
1606 struct kvm_s390_vm_cpu_processor *proc;
1607 int ret = 0;
1608
1609 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1610 if (!proc) {
1611 ret = -ENOMEM;
1612 goto out;
1613 }
1614 proc->cpuid = kvm->arch.model.cpuid;
1615 proc->ibc = kvm->arch.model.ibc;
1616 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1617 S390_ARCH_FAC_LIST_SIZE_BYTE);
1618 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1619 kvm->arch.model.ibc,
1620 kvm->arch.model.cpuid);
1621 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1622 kvm->arch.model.fac_list[0],
1623 kvm->arch.model.fac_list[1],
1624 kvm->arch.model.fac_list[2]);
1625 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1626 ret = -EFAULT;
1627 kfree(proc);
1628 out:
1629 return ret;
1630 }
1631
kvm_s390_get_machine(struct kvm * kvm,struct kvm_device_attr * attr)1632 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1633 {
1634 struct kvm_s390_vm_cpu_machine *mach;
1635 int ret = 0;
1636
1637 mach = kzalloc(sizeof(*mach), GFP_KERNEL_ACCOUNT);
1638 if (!mach) {
1639 ret = -ENOMEM;
1640 goto out;
1641 }
1642 get_cpu_id((struct cpuid *) &mach->cpuid);
1643 mach->ibc = sclp.ibc;
1644 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1645 S390_ARCH_FAC_LIST_SIZE_BYTE);
1646 memcpy((unsigned long *)&mach->fac_list, stfle_fac_list,
1647 sizeof(stfle_fac_list));
1648 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
1649 kvm->arch.model.ibc,
1650 kvm->arch.model.cpuid);
1651 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
1652 mach->fac_mask[0],
1653 mach->fac_mask[1],
1654 mach->fac_mask[2]);
1655 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1656 mach->fac_list[0],
1657 mach->fac_list[1],
1658 mach->fac_list[2]);
1659 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1660 ret = -EFAULT;
1661 kfree(mach);
1662 out:
1663 return ret;
1664 }
1665
kvm_s390_get_processor_feat(struct kvm * kvm,struct kvm_device_attr * attr)1666 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1667 struct kvm_device_attr *attr)
1668 {
1669 struct kvm_s390_vm_cpu_feat data;
1670
1671 bitmap_to_arr64(data.feat, kvm->arch.cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1672 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1673 return -EFAULT;
1674 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1675 data.feat[0],
1676 data.feat[1],
1677 data.feat[2]);
1678 return 0;
1679 }
1680
kvm_s390_get_machine_feat(struct kvm * kvm,struct kvm_device_attr * attr)1681 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1682 struct kvm_device_attr *attr)
1683 {
1684 struct kvm_s390_vm_cpu_feat data;
1685
1686 bitmap_to_arr64(data.feat, kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1687 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1688 return -EFAULT;
1689 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1690 data.feat[0],
1691 data.feat[1],
1692 data.feat[2]);
1693 return 0;
1694 }
1695
kvm_s390_get_processor_subfunc(struct kvm * kvm,struct kvm_device_attr * attr)1696 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1697 struct kvm_device_attr *attr)
1698 {
1699 if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs,
1700 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1701 return -EFAULT;
1702
1703 VM_EVENT(kvm, 3, "GET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1704 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1705 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1706 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1707 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1708 VM_EVENT(kvm, 3, "GET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1709 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1710 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1711 VM_EVENT(kvm, 3, "GET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1712 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1713 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1714 VM_EVENT(kvm, 3, "GET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1715 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1716 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1717 VM_EVENT(kvm, 3, "GET: guest KM subfunc 0x%16.16lx.%16.16lx",
1718 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1719 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1720 VM_EVENT(kvm, 3, "GET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1721 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1722 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1723 VM_EVENT(kvm, 3, "GET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1724 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1725 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1726 VM_EVENT(kvm, 3, "GET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1727 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1728 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1729 VM_EVENT(kvm, 3, "GET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1730 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1731 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1732 VM_EVENT(kvm, 3, "GET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1733 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1734 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1735 VM_EVENT(kvm, 3, "GET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1736 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1737 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1738 VM_EVENT(kvm, 3, "GET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1739 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1740 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1741 VM_EVENT(kvm, 3, "GET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1742 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1743 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1744 VM_EVENT(kvm, 3, "GET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1745 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1746 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1747 VM_EVENT(kvm, 3, "GET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1748 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1749 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1750 VM_EVENT(kvm, 3, "GET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1751 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1752 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1753 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1754 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1755 VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1756 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1757 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1758 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1759 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1760
1761 return 0;
1762 }
1763
kvm_s390_get_machine_subfunc(struct kvm * kvm,struct kvm_device_attr * attr)1764 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1765 struct kvm_device_attr *attr)
1766 {
1767 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1768 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1769 return -EFAULT;
1770
1771 VM_EVENT(kvm, 3, "GET: host PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1772 ((unsigned long *) &kvm_s390_available_subfunc.plo)[0],
1773 ((unsigned long *) &kvm_s390_available_subfunc.plo)[1],
1774 ((unsigned long *) &kvm_s390_available_subfunc.plo)[2],
1775 ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]);
1776 VM_EVENT(kvm, 3, "GET: host PTFF subfunc 0x%16.16lx.%16.16lx",
1777 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0],
1778 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]);
1779 VM_EVENT(kvm, 3, "GET: host KMAC subfunc 0x%16.16lx.%16.16lx",
1780 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0],
1781 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]);
1782 VM_EVENT(kvm, 3, "GET: host KMC subfunc 0x%16.16lx.%16.16lx",
1783 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0],
1784 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]);
1785 VM_EVENT(kvm, 3, "GET: host KM subfunc 0x%16.16lx.%16.16lx",
1786 ((unsigned long *) &kvm_s390_available_subfunc.km)[0],
1787 ((unsigned long *) &kvm_s390_available_subfunc.km)[1]);
1788 VM_EVENT(kvm, 3, "GET: host KIMD subfunc 0x%16.16lx.%16.16lx",
1789 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0],
1790 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]);
1791 VM_EVENT(kvm, 3, "GET: host KLMD subfunc 0x%16.16lx.%16.16lx",
1792 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0],
1793 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]);
1794 VM_EVENT(kvm, 3, "GET: host PCKMO subfunc 0x%16.16lx.%16.16lx",
1795 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0],
1796 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]);
1797 VM_EVENT(kvm, 3, "GET: host KMCTR subfunc 0x%16.16lx.%16.16lx",
1798 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0],
1799 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]);
1800 VM_EVENT(kvm, 3, "GET: host KMF subfunc 0x%16.16lx.%16.16lx",
1801 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0],
1802 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]);
1803 VM_EVENT(kvm, 3, "GET: host KMO subfunc 0x%16.16lx.%16.16lx",
1804 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0],
1805 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]);
1806 VM_EVENT(kvm, 3, "GET: host PCC subfunc 0x%16.16lx.%16.16lx",
1807 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0],
1808 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]);
1809 VM_EVENT(kvm, 3, "GET: host PPNO subfunc 0x%16.16lx.%16.16lx",
1810 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0],
1811 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]);
1812 VM_EVENT(kvm, 3, "GET: host KMA subfunc 0x%16.16lx.%16.16lx",
1813 ((unsigned long *) &kvm_s390_available_subfunc.kma)[0],
1814 ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]);
1815 VM_EVENT(kvm, 3, "GET: host KDSA subfunc 0x%16.16lx.%16.16lx",
1816 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0],
1817 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]);
1818 VM_EVENT(kvm, 3, "GET: host SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1819 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0],
1820 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1],
1821 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2],
1822 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]);
1823 VM_EVENT(kvm, 3, "GET: host DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1824 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0],
1825 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1],
1826 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2],
1827 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]);
1828
1829 return 0;
1830 }
1831
kvm_s390_get_processor_uv_feat(struct kvm * kvm,struct kvm_device_attr * attr)1832 static int kvm_s390_get_processor_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1833 {
1834 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1835 unsigned long feat = kvm->arch.model.uv_feat_guest.feat;
1836
1837 if (put_user(feat, &dst->feat))
1838 return -EFAULT;
1839 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1840
1841 return 0;
1842 }
1843
kvm_s390_get_machine_uv_feat(struct kvm * kvm,struct kvm_device_attr * attr)1844 static int kvm_s390_get_machine_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1845 {
1846 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1847 unsigned long feat;
1848
1849 BUILD_BUG_ON(sizeof(*dst) != sizeof(uv_info.uv_feature_indications));
1850
1851 feat = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1852 if (put_user(feat, &dst->feat))
1853 return -EFAULT;
1854 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1855
1856 return 0;
1857 }
1858
kvm_s390_get_cpu_model(struct kvm * kvm,struct kvm_device_attr * attr)1859 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1860 {
1861 int ret = -ENXIO;
1862
1863 switch (attr->attr) {
1864 case KVM_S390_VM_CPU_PROCESSOR:
1865 ret = kvm_s390_get_processor(kvm, attr);
1866 break;
1867 case KVM_S390_VM_CPU_MACHINE:
1868 ret = kvm_s390_get_machine(kvm, attr);
1869 break;
1870 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1871 ret = kvm_s390_get_processor_feat(kvm, attr);
1872 break;
1873 case KVM_S390_VM_CPU_MACHINE_FEAT:
1874 ret = kvm_s390_get_machine_feat(kvm, attr);
1875 break;
1876 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1877 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1878 break;
1879 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1880 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1881 break;
1882 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1883 ret = kvm_s390_get_processor_uv_feat(kvm, attr);
1884 break;
1885 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
1886 ret = kvm_s390_get_machine_uv_feat(kvm, attr);
1887 break;
1888 }
1889 return ret;
1890 }
1891
1892 /**
1893 * kvm_s390_update_topology_change_report - update CPU topology change report
1894 * @kvm: guest KVM description
1895 * @val: set or clear the MTCR bit
1896 *
1897 * Updates the Multiprocessor Topology-Change-Report bit to signal
1898 * the guest with a topology change.
1899 * This is only relevant if the topology facility is present.
1900 *
1901 * The SCA version, bsca or esca, doesn't matter as offset is the same.
1902 */
kvm_s390_update_topology_change_report(struct kvm * kvm,bool val)1903 static void kvm_s390_update_topology_change_report(struct kvm *kvm, bool val)
1904 {
1905 union sca_utility new, old;
1906 struct bsca_block *sca;
1907
1908 read_lock(&kvm->arch.sca_lock);
1909 sca = kvm->arch.sca;
1910 do {
1911 old = READ_ONCE(sca->utility);
1912 new = old;
1913 new.mtcr = val;
1914 } while (cmpxchg(&sca->utility.val, old.val, new.val) != old.val);
1915 read_unlock(&kvm->arch.sca_lock);
1916 }
1917
kvm_s390_set_topo_change_indication(struct kvm * kvm,struct kvm_device_attr * attr)1918 static int kvm_s390_set_topo_change_indication(struct kvm *kvm,
1919 struct kvm_device_attr *attr)
1920 {
1921 if (!test_kvm_facility(kvm, 11))
1922 return -ENXIO;
1923
1924 kvm_s390_update_topology_change_report(kvm, !!attr->attr);
1925 return 0;
1926 }
1927
kvm_s390_get_topo_change_indication(struct kvm * kvm,struct kvm_device_attr * attr)1928 static int kvm_s390_get_topo_change_indication(struct kvm *kvm,
1929 struct kvm_device_attr *attr)
1930 {
1931 u8 topo;
1932
1933 if (!test_kvm_facility(kvm, 11))
1934 return -ENXIO;
1935
1936 read_lock(&kvm->arch.sca_lock);
1937 topo = ((struct bsca_block *)kvm->arch.sca)->utility.mtcr;
1938 read_unlock(&kvm->arch.sca_lock);
1939
1940 return put_user(topo, (u8 __user *)attr->addr);
1941 }
1942
kvm_s390_vm_set_attr(struct kvm * kvm,struct kvm_device_attr * attr)1943 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1944 {
1945 int ret;
1946
1947 switch (attr->group) {
1948 case KVM_S390_VM_MEM_CTRL:
1949 ret = kvm_s390_set_mem_control(kvm, attr);
1950 break;
1951 case KVM_S390_VM_TOD:
1952 ret = kvm_s390_set_tod(kvm, attr);
1953 break;
1954 case KVM_S390_VM_CPU_MODEL:
1955 ret = kvm_s390_set_cpu_model(kvm, attr);
1956 break;
1957 case KVM_S390_VM_CRYPTO:
1958 ret = kvm_s390_vm_set_crypto(kvm, attr);
1959 break;
1960 case KVM_S390_VM_MIGRATION:
1961 ret = kvm_s390_vm_set_migration(kvm, attr);
1962 break;
1963 case KVM_S390_VM_CPU_TOPOLOGY:
1964 ret = kvm_s390_set_topo_change_indication(kvm, attr);
1965 break;
1966 default:
1967 ret = -ENXIO;
1968 break;
1969 }
1970
1971 return ret;
1972 }
1973
kvm_s390_vm_get_attr(struct kvm * kvm,struct kvm_device_attr * attr)1974 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1975 {
1976 int ret;
1977
1978 switch (attr->group) {
1979 case KVM_S390_VM_MEM_CTRL:
1980 ret = kvm_s390_get_mem_control(kvm, attr);
1981 break;
1982 case KVM_S390_VM_TOD:
1983 ret = kvm_s390_get_tod(kvm, attr);
1984 break;
1985 case KVM_S390_VM_CPU_MODEL:
1986 ret = kvm_s390_get_cpu_model(kvm, attr);
1987 break;
1988 case KVM_S390_VM_MIGRATION:
1989 ret = kvm_s390_vm_get_migration(kvm, attr);
1990 break;
1991 case KVM_S390_VM_CPU_TOPOLOGY:
1992 ret = kvm_s390_get_topo_change_indication(kvm, attr);
1993 break;
1994 default:
1995 ret = -ENXIO;
1996 break;
1997 }
1998
1999 return ret;
2000 }
2001
kvm_s390_vm_has_attr(struct kvm * kvm,struct kvm_device_attr * attr)2002 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
2003 {
2004 int ret;
2005
2006 switch (attr->group) {
2007 case KVM_S390_VM_MEM_CTRL:
2008 switch (attr->attr) {
2009 case KVM_S390_VM_MEM_ENABLE_CMMA:
2010 case KVM_S390_VM_MEM_CLR_CMMA:
2011 ret = sclp.has_cmma ? 0 : -ENXIO;
2012 break;
2013 case KVM_S390_VM_MEM_LIMIT_SIZE:
2014 ret = 0;
2015 break;
2016 default:
2017 ret = -ENXIO;
2018 break;
2019 }
2020 break;
2021 case KVM_S390_VM_TOD:
2022 switch (attr->attr) {
2023 case KVM_S390_VM_TOD_LOW:
2024 case KVM_S390_VM_TOD_HIGH:
2025 ret = 0;
2026 break;
2027 default:
2028 ret = -ENXIO;
2029 break;
2030 }
2031 break;
2032 case KVM_S390_VM_CPU_MODEL:
2033 switch (attr->attr) {
2034 case KVM_S390_VM_CPU_PROCESSOR:
2035 case KVM_S390_VM_CPU_MACHINE:
2036 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
2037 case KVM_S390_VM_CPU_MACHINE_FEAT:
2038 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
2039 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
2040 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
2041 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
2042 ret = 0;
2043 break;
2044 default:
2045 ret = -ENXIO;
2046 break;
2047 }
2048 break;
2049 case KVM_S390_VM_CRYPTO:
2050 switch (attr->attr) {
2051 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
2052 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
2053 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
2054 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
2055 ret = 0;
2056 break;
2057 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
2058 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
2059 ret = ap_instructions_available() ? 0 : -ENXIO;
2060 break;
2061 default:
2062 ret = -ENXIO;
2063 break;
2064 }
2065 break;
2066 case KVM_S390_VM_MIGRATION:
2067 ret = 0;
2068 break;
2069 case KVM_S390_VM_CPU_TOPOLOGY:
2070 ret = test_kvm_facility(kvm, 11) ? 0 : -ENXIO;
2071 break;
2072 default:
2073 ret = -ENXIO;
2074 break;
2075 }
2076
2077 return ret;
2078 }
2079
kvm_s390_get_skeys(struct kvm * kvm,struct kvm_s390_skeys * args)2080 static int kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2081 {
2082 uint8_t *keys;
2083 uint64_t hva;
2084 int srcu_idx, i, r = 0;
2085
2086 if (args->flags != 0)
2087 return -EINVAL;
2088
2089 /* Is this guest using storage keys? */
2090 if (!mm_uses_skeys(current->mm))
2091 return KVM_S390_GET_SKEYS_NONE;
2092
2093 /* Enforce sane limit on memory allocation */
2094 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2095 return -EINVAL;
2096
2097 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2098 if (!keys)
2099 return -ENOMEM;
2100
2101 mmap_read_lock(current->mm);
2102 srcu_idx = srcu_read_lock(&kvm->srcu);
2103 for (i = 0; i < args->count; i++) {
2104 hva = gfn_to_hva(kvm, args->start_gfn + i);
2105 if (kvm_is_error_hva(hva)) {
2106 r = -EFAULT;
2107 break;
2108 }
2109
2110 r = get_guest_storage_key(current->mm, hva, &keys[i]);
2111 if (r)
2112 break;
2113 }
2114 srcu_read_unlock(&kvm->srcu, srcu_idx);
2115 mmap_read_unlock(current->mm);
2116
2117 if (!r) {
2118 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
2119 sizeof(uint8_t) * args->count);
2120 if (r)
2121 r = -EFAULT;
2122 }
2123
2124 kvfree(keys);
2125 return r;
2126 }
2127
kvm_s390_set_skeys(struct kvm * kvm,struct kvm_s390_skeys * args)2128 static int kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2129 {
2130 uint8_t *keys;
2131 uint64_t hva;
2132 int srcu_idx, i, r = 0;
2133 bool unlocked;
2134
2135 if (args->flags != 0)
2136 return -EINVAL;
2137
2138 /* Enforce sane limit on memory allocation */
2139 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2140 return -EINVAL;
2141
2142 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2143 if (!keys)
2144 return -ENOMEM;
2145
2146 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
2147 sizeof(uint8_t) * args->count);
2148 if (r) {
2149 r = -EFAULT;
2150 goto out;
2151 }
2152
2153 /* Enable storage key handling for the guest */
2154 r = s390_enable_skey();
2155 if (r)
2156 goto out;
2157
2158 i = 0;
2159 mmap_read_lock(current->mm);
2160 srcu_idx = srcu_read_lock(&kvm->srcu);
2161 while (i < args->count) {
2162 unlocked = false;
2163 hva = gfn_to_hva(kvm, args->start_gfn + i);
2164 if (kvm_is_error_hva(hva)) {
2165 r = -EFAULT;
2166 break;
2167 }
2168
2169 /* Lowest order bit is reserved */
2170 if (keys[i] & 0x01) {
2171 r = -EINVAL;
2172 break;
2173 }
2174
2175 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
2176 if (r) {
2177 r = fixup_user_fault(current->mm, hva,
2178 FAULT_FLAG_WRITE, &unlocked);
2179 if (r)
2180 break;
2181 }
2182 if (!r)
2183 i++;
2184 }
2185 srcu_read_unlock(&kvm->srcu, srcu_idx);
2186 mmap_read_unlock(current->mm);
2187 out:
2188 kvfree(keys);
2189 return r;
2190 }
2191
2192 /*
2193 * Base address and length must be sent at the start of each block, therefore
2194 * it's cheaper to send some clean data, as long as it's less than the size of
2195 * two longs.
2196 */
2197 #define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
2198 /* for consistency */
2199 #define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
2200
kvm_s390_peek_cmma(struct kvm * kvm,struct kvm_s390_cmma_log * args,u8 * res,unsigned long bufsize)2201 static int kvm_s390_peek_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2202 u8 *res, unsigned long bufsize)
2203 {
2204 unsigned long pgstev, hva, cur_gfn = args->start_gfn;
2205
2206 args->count = 0;
2207 while (args->count < bufsize) {
2208 hva = gfn_to_hva(kvm, cur_gfn);
2209 /*
2210 * We return an error if the first value was invalid, but we
2211 * return successfully if at least one value was copied.
2212 */
2213 if (kvm_is_error_hva(hva))
2214 return args->count ? 0 : -EFAULT;
2215 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2216 pgstev = 0;
2217 res[args->count++] = (pgstev >> 24) & 0x43;
2218 cur_gfn++;
2219 }
2220
2221 return 0;
2222 }
2223
gfn_to_memslot_approx(struct kvm_memslots * slots,gfn_t gfn)2224 static struct kvm_memory_slot *gfn_to_memslot_approx(struct kvm_memslots *slots,
2225 gfn_t gfn)
2226 {
2227 return ____gfn_to_memslot(slots, gfn, true);
2228 }
2229
kvm_s390_next_dirty_cmma(struct kvm_memslots * slots,unsigned long cur_gfn)2230 static unsigned long kvm_s390_next_dirty_cmma(struct kvm_memslots *slots,
2231 unsigned long cur_gfn)
2232 {
2233 struct kvm_memory_slot *ms = gfn_to_memslot_approx(slots, cur_gfn);
2234 unsigned long ofs = cur_gfn - ms->base_gfn;
2235 struct rb_node *mnode = &ms->gfn_node[slots->node_idx];
2236
2237 if (ms->base_gfn + ms->npages <= cur_gfn) {
2238 mnode = rb_next(mnode);
2239 /* If we are above the highest slot, wrap around */
2240 if (!mnode)
2241 mnode = rb_first(&slots->gfn_tree);
2242
2243 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2244 ofs = 0;
2245 }
2246
2247 if (cur_gfn < ms->base_gfn)
2248 ofs = 0;
2249
2250 ofs = find_next_bit(kvm_second_dirty_bitmap(ms), ms->npages, ofs);
2251 while (ofs >= ms->npages && (mnode = rb_next(mnode))) {
2252 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2253 ofs = find_first_bit(kvm_second_dirty_bitmap(ms), ms->npages);
2254 }
2255 return ms->base_gfn + ofs;
2256 }
2257
kvm_s390_get_cmma(struct kvm * kvm,struct kvm_s390_cmma_log * args,u8 * res,unsigned long bufsize)2258 static int kvm_s390_get_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2259 u8 *res, unsigned long bufsize)
2260 {
2261 unsigned long mem_end, cur_gfn, next_gfn, hva, pgstev;
2262 struct kvm_memslots *slots = kvm_memslots(kvm);
2263 struct kvm_memory_slot *ms;
2264
2265 if (unlikely(kvm_memslots_empty(slots)))
2266 return 0;
2267
2268 cur_gfn = kvm_s390_next_dirty_cmma(slots, args->start_gfn);
2269 ms = gfn_to_memslot(kvm, cur_gfn);
2270 args->count = 0;
2271 args->start_gfn = cur_gfn;
2272 if (!ms)
2273 return 0;
2274 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2275 mem_end = kvm_s390_get_gfn_end(slots);
2276
2277 while (args->count < bufsize) {
2278 hva = gfn_to_hva(kvm, cur_gfn);
2279 if (kvm_is_error_hva(hva))
2280 return 0;
2281 /* Decrement only if we actually flipped the bit to 0 */
2282 if (test_and_clear_bit(cur_gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
2283 atomic64_dec(&kvm->arch.cmma_dirty_pages);
2284 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2285 pgstev = 0;
2286 /* Save the value */
2287 res[args->count++] = (pgstev >> 24) & 0x43;
2288 /* If the next bit is too far away, stop. */
2289 if (next_gfn > cur_gfn + KVM_S390_MAX_BIT_DISTANCE)
2290 return 0;
2291 /* If we reached the previous "next", find the next one */
2292 if (cur_gfn == next_gfn)
2293 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2294 /* Reached the end of memory or of the buffer, stop */
2295 if ((next_gfn >= mem_end) ||
2296 (next_gfn - args->start_gfn >= bufsize))
2297 return 0;
2298 cur_gfn++;
2299 /* Reached the end of the current memslot, take the next one. */
2300 if (cur_gfn - ms->base_gfn >= ms->npages) {
2301 ms = gfn_to_memslot(kvm, cur_gfn);
2302 if (!ms)
2303 return 0;
2304 }
2305 }
2306 return 0;
2307 }
2308
2309 /*
2310 * This function searches for the next page with dirty CMMA attributes, and
2311 * saves the attributes in the buffer up to either the end of the buffer or
2312 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
2313 * no trailing clean bytes are saved.
2314 * In case no dirty bits were found, or if CMMA was not enabled or used, the
2315 * output buffer will indicate 0 as length.
2316 */
kvm_s390_get_cmma_bits(struct kvm * kvm,struct kvm_s390_cmma_log * args)2317 static int kvm_s390_get_cmma_bits(struct kvm *kvm,
2318 struct kvm_s390_cmma_log *args)
2319 {
2320 unsigned long bufsize;
2321 int srcu_idx, peek, ret;
2322 u8 *values;
2323
2324 if (!kvm->arch.use_cmma)
2325 return -ENXIO;
2326 /* Invalid/unsupported flags were specified */
2327 if (args->flags & ~KVM_S390_CMMA_PEEK)
2328 return -EINVAL;
2329 /* Migration mode query, and we are not doing a migration */
2330 peek = !!(args->flags & KVM_S390_CMMA_PEEK);
2331 if (!peek && !kvm->arch.migration_mode)
2332 return -EINVAL;
2333 /* CMMA is disabled or was not used, or the buffer has length zero */
2334 bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
2335 if (!bufsize || !kvm->mm->context.uses_cmm) {
2336 memset(args, 0, sizeof(*args));
2337 return 0;
2338 }
2339 /* We are not peeking, and there are no dirty pages */
2340 if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) {
2341 memset(args, 0, sizeof(*args));
2342 return 0;
2343 }
2344
2345 values = vmalloc(bufsize);
2346 if (!values)
2347 return -ENOMEM;
2348
2349 mmap_read_lock(kvm->mm);
2350 srcu_idx = srcu_read_lock(&kvm->srcu);
2351 if (peek)
2352 ret = kvm_s390_peek_cmma(kvm, args, values, bufsize);
2353 else
2354 ret = kvm_s390_get_cmma(kvm, args, values, bufsize);
2355 srcu_read_unlock(&kvm->srcu, srcu_idx);
2356 mmap_read_unlock(kvm->mm);
2357
2358 if (kvm->arch.migration_mode)
2359 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages);
2360 else
2361 args->remaining = 0;
2362
2363 if (copy_to_user((void __user *)args->values, values, args->count))
2364 ret = -EFAULT;
2365
2366 vfree(values);
2367 return ret;
2368 }
2369
2370 /*
2371 * This function sets the CMMA attributes for the given pages. If the input
2372 * buffer has zero length, no action is taken, otherwise the attributes are
2373 * set and the mm->context.uses_cmm flag is set.
2374 */
kvm_s390_set_cmma_bits(struct kvm * kvm,const struct kvm_s390_cmma_log * args)2375 static int kvm_s390_set_cmma_bits(struct kvm *kvm,
2376 const struct kvm_s390_cmma_log *args)
2377 {
2378 unsigned long hva, mask, pgstev, i;
2379 uint8_t *bits;
2380 int srcu_idx, r = 0;
2381
2382 mask = args->mask;
2383
2384 if (!kvm->arch.use_cmma)
2385 return -ENXIO;
2386 /* invalid/unsupported flags */
2387 if (args->flags != 0)
2388 return -EINVAL;
2389 /* Enforce sane limit on memory allocation */
2390 if (args->count > KVM_S390_CMMA_SIZE_MAX)
2391 return -EINVAL;
2392 /* Nothing to do */
2393 if (args->count == 0)
2394 return 0;
2395
2396 bits = vmalloc(array_size(sizeof(*bits), args->count));
2397 if (!bits)
2398 return -ENOMEM;
2399
2400 r = copy_from_user(bits, (void __user *)args->values, args->count);
2401 if (r) {
2402 r = -EFAULT;
2403 goto out;
2404 }
2405
2406 mmap_read_lock(kvm->mm);
2407 srcu_idx = srcu_read_lock(&kvm->srcu);
2408 for (i = 0; i < args->count; i++) {
2409 hva = gfn_to_hva(kvm, args->start_gfn + i);
2410 if (kvm_is_error_hva(hva)) {
2411 r = -EFAULT;
2412 break;
2413 }
2414
2415 pgstev = bits[i];
2416 pgstev = pgstev << 24;
2417 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
2418 set_pgste_bits(kvm->mm, hva, mask, pgstev);
2419 }
2420 srcu_read_unlock(&kvm->srcu, srcu_idx);
2421 mmap_read_unlock(kvm->mm);
2422
2423 if (!kvm->mm->context.uses_cmm) {
2424 mmap_write_lock(kvm->mm);
2425 kvm->mm->context.uses_cmm = 1;
2426 mmap_write_unlock(kvm->mm);
2427 }
2428 out:
2429 vfree(bits);
2430 return r;
2431 }
2432
2433 /**
2434 * kvm_s390_cpus_from_pv - Convert all protected vCPUs in a protected VM to
2435 * non protected.
2436 * @kvm: the VM whose protected vCPUs are to be converted
2437 * @rc: return value for the RC field of the UVC (in case of error)
2438 * @rrc: return value for the RRC field of the UVC (in case of error)
2439 *
2440 * Does not stop in case of error, tries to convert as many
2441 * CPUs as possible. In case of error, the RC and RRC of the last error are
2442 * returned.
2443 *
2444 * Return: 0 in case of success, otherwise -EIO
2445 */
kvm_s390_cpus_from_pv(struct kvm * kvm,u16 * rc,u16 * rrc)2446 int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2447 {
2448 struct kvm_vcpu *vcpu;
2449 unsigned long i;
2450 u16 _rc, _rrc;
2451 int ret = 0;
2452
2453 /*
2454 * We ignore failures and try to destroy as many CPUs as possible.
2455 * At the same time we must not free the assigned resources when
2456 * this fails, as the ultravisor has still access to that memory.
2457 * So kvm_s390_pv_destroy_cpu can leave a "wanted" memory leak
2458 * behind.
2459 * We want to return the first failure rc and rrc, though.
2460 */
2461 kvm_for_each_vcpu(i, vcpu, kvm) {
2462 mutex_lock(&vcpu->mutex);
2463 if (kvm_s390_pv_destroy_cpu(vcpu, &_rc, &_rrc) && !ret) {
2464 *rc = _rc;
2465 *rrc = _rrc;
2466 ret = -EIO;
2467 }
2468 mutex_unlock(&vcpu->mutex);
2469 }
2470 /* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */
2471 if (use_gisa)
2472 kvm_s390_gisa_enable(kvm);
2473 return ret;
2474 }
2475
2476 /**
2477 * kvm_s390_cpus_to_pv - Convert all non-protected vCPUs in a protected VM
2478 * to protected.
2479 * @kvm: the VM whose protected vCPUs are to be converted
2480 * @rc: return value for the RC field of the UVC (in case of error)
2481 * @rrc: return value for the RRC field of the UVC (in case of error)
2482 *
2483 * Tries to undo the conversion in case of error.
2484 *
2485 * Return: 0 in case of success, otherwise -EIO
2486 */
kvm_s390_cpus_to_pv(struct kvm * kvm,u16 * rc,u16 * rrc)2487 static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2488 {
2489 unsigned long i;
2490 int r = 0;
2491 u16 dummy;
2492
2493 struct kvm_vcpu *vcpu;
2494
2495 /* Disable the GISA if the ultravisor does not support AIV. */
2496 if (!uv_has_feature(BIT_UV_FEAT_AIV))
2497 kvm_s390_gisa_disable(kvm);
2498
2499 kvm_for_each_vcpu(i, vcpu, kvm) {
2500 mutex_lock(&vcpu->mutex);
2501 r = kvm_s390_pv_create_cpu(vcpu, rc, rrc);
2502 mutex_unlock(&vcpu->mutex);
2503 if (r)
2504 break;
2505 }
2506 if (r)
2507 kvm_s390_cpus_from_pv(kvm, &dummy, &dummy);
2508 return r;
2509 }
2510
2511 /*
2512 * Here we provide user space with a direct interface to query UV
2513 * related data like UV maxima and available features as well as
2514 * feature specific data.
2515 *
2516 * To facilitate future extension of the data structures we'll try to
2517 * write data up to the maximum requested length.
2518 */
kvm_s390_handle_pv_info(struct kvm_s390_pv_info * info)2519 static ssize_t kvm_s390_handle_pv_info(struct kvm_s390_pv_info *info)
2520 {
2521 ssize_t len_min;
2522
2523 switch (info->header.id) {
2524 case KVM_PV_INFO_VM: {
2525 len_min = sizeof(info->header) + sizeof(info->vm);
2526
2527 if (info->header.len_max < len_min)
2528 return -EINVAL;
2529
2530 memcpy(info->vm.inst_calls_list,
2531 uv_info.inst_calls_list,
2532 sizeof(uv_info.inst_calls_list));
2533
2534 /* It's max cpuid not max cpus, so it's off by one */
2535 info->vm.max_cpus = uv_info.max_guest_cpu_id + 1;
2536 info->vm.max_guests = uv_info.max_num_sec_conf;
2537 info->vm.max_guest_addr = uv_info.max_sec_stor_addr;
2538 info->vm.feature_indication = uv_info.uv_feature_indications;
2539
2540 return len_min;
2541 }
2542 case KVM_PV_INFO_DUMP: {
2543 len_min = sizeof(info->header) + sizeof(info->dump);
2544
2545 if (info->header.len_max < len_min)
2546 return -EINVAL;
2547
2548 info->dump.dump_cpu_buffer_len = uv_info.guest_cpu_stor_len;
2549 info->dump.dump_config_mem_buffer_per_1m = uv_info.conf_dump_storage_state_len;
2550 info->dump.dump_config_finalize_len = uv_info.conf_dump_finalize_len;
2551 return len_min;
2552 }
2553 default:
2554 return -EINVAL;
2555 }
2556 }
2557
kvm_s390_pv_dmp(struct kvm * kvm,struct kvm_pv_cmd * cmd,struct kvm_s390_pv_dmp dmp)2558 static int kvm_s390_pv_dmp(struct kvm *kvm, struct kvm_pv_cmd *cmd,
2559 struct kvm_s390_pv_dmp dmp)
2560 {
2561 int r = -EINVAL;
2562 void __user *result_buff = (void __user *)dmp.buff_addr;
2563
2564 switch (dmp.subcmd) {
2565 case KVM_PV_DUMP_INIT: {
2566 if (kvm->arch.pv.dumping)
2567 break;
2568
2569 /*
2570 * Block SIE entry as concurrent dump UVCs could lead
2571 * to validities.
2572 */
2573 kvm_s390_vcpu_block_all(kvm);
2574
2575 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2576 UVC_CMD_DUMP_INIT, &cmd->rc, &cmd->rrc);
2577 KVM_UV_EVENT(kvm, 3, "PROTVIRT DUMP INIT: rc %x rrc %x",
2578 cmd->rc, cmd->rrc);
2579 if (!r) {
2580 kvm->arch.pv.dumping = true;
2581 } else {
2582 kvm_s390_vcpu_unblock_all(kvm);
2583 r = -EINVAL;
2584 }
2585 break;
2586 }
2587 case KVM_PV_DUMP_CONFIG_STOR_STATE: {
2588 if (!kvm->arch.pv.dumping)
2589 break;
2590
2591 /*
2592 * gaddr is an output parameter since we might stop
2593 * early. As dmp will be copied back in our caller, we
2594 * don't need to do it ourselves.
2595 */
2596 r = kvm_s390_pv_dump_stor_state(kvm, result_buff, &dmp.gaddr, dmp.buff_len,
2597 &cmd->rc, &cmd->rrc);
2598 break;
2599 }
2600 case KVM_PV_DUMP_COMPLETE: {
2601 if (!kvm->arch.pv.dumping)
2602 break;
2603
2604 r = -EINVAL;
2605 if (dmp.buff_len < uv_info.conf_dump_finalize_len)
2606 break;
2607
2608 r = kvm_s390_pv_dump_complete(kvm, result_buff,
2609 &cmd->rc, &cmd->rrc);
2610 break;
2611 }
2612 default:
2613 r = -ENOTTY;
2614 break;
2615 }
2616
2617 return r;
2618 }
2619
kvm_s390_handle_pv(struct kvm * kvm,struct kvm_pv_cmd * cmd)2620 static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd)
2621 {
2622 const bool need_lock = (cmd->cmd != KVM_PV_ASYNC_CLEANUP_PERFORM);
2623 void __user *argp = (void __user *)cmd->data;
2624 int r = 0;
2625 u16 dummy;
2626
2627 if (need_lock)
2628 mutex_lock(&kvm->lock);
2629
2630 switch (cmd->cmd) {
2631 case KVM_PV_ENABLE: {
2632 r = -EINVAL;
2633 if (kvm_s390_pv_is_protected(kvm))
2634 break;
2635
2636 /*
2637 * FMT 4 SIE needs esca. As we never switch back to bsca from
2638 * esca, we need no cleanup in the error cases below
2639 */
2640 r = sca_switch_to_extended(kvm);
2641 if (r)
2642 break;
2643
2644 r = s390_disable_cow_sharing();
2645 if (r)
2646 break;
2647
2648 r = kvm_s390_pv_init_vm(kvm, &cmd->rc, &cmd->rrc);
2649 if (r)
2650 break;
2651
2652 r = kvm_s390_cpus_to_pv(kvm, &cmd->rc, &cmd->rrc);
2653 if (r)
2654 kvm_s390_pv_deinit_vm(kvm, &dummy, &dummy);
2655
2656 /* we need to block service interrupts from now on */
2657 set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2658 break;
2659 }
2660 case KVM_PV_ASYNC_CLEANUP_PREPARE:
2661 r = -EINVAL;
2662 if (!kvm_s390_pv_is_protected(kvm) || !async_destroy)
2663 break;
2664
2665 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2666 /*
2667 * If a CPU could not be destroyed, destroy VM will also fail.
2668 * There is no point in trying to destroy it. Instead return
2669 * the rc and rrc from the first CPU that failed destroying.
2670 */
2671 if (r)
2672 break;
2673 r = kvm_s390_pv_set_aside(kvm, &cmd->rc, &cmd->rrc);
2674
2675 /* no need to block service interrupts any more */
2676 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2677 break;
2678 case KVM_PV_ASYNC_CLEANUP_PERFORM:
2679 r = -EINVAL;
2680 if (!async_destroy)
2681 break;
2682 /* kvm->lock must not be held; this is asserted inside the function. */
2683 r = kvm_s390_pv_deinit_aside_vm(kvm, &cmd->rc, &cmd->rrc);
2684 break;
2685 case KVM_PV_DISABLE: {
2686 r = -EINVAL;
2687 if (!kvm_s390_pv_is_protected(kvm))
2688 break;
2689
2690 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2691 /*
2692 * If a CPU could not be destroyed, destroy VM will also fail.
2693 * There is no point in trying to destroy it. Instead return
2694 * the rc and rrc from the first CPU that failed destroying.
2695 */
2696 if (r)
2697 break;
2698 r = kvm_s390_pv_deinit_cleanup_all(kvm, &cmd->rc, &cmd->rrc);
2699
2700 /* no need to block service interrupts any more */
2701 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2702 break;
2703 }
2704 case KVM_PV_SET_SEC_PARMS: {
2705 struct kvm_s390_pv_sec_parm parms = {};
2706 void *hdr;
2707
2708 r = -EINVAL;
2709 if (!kvm_s390_pv_is_protected(kvm))
2710 break;
2711
2712 r = -EFAULT;
2713 if (copy_from_user(&parms, argp, sizeof(parms)))
2714 break;
2715
2716 /* Currently restricted to 8KB */
2717 r = -EINVAL;
2718 if (parms.length > PAGE_SIZE * 2)
2719 break;
2720
2721 r = -ENOMEM;
2722 hdr = vmalloc(parms.length);
2723 if (!hdr)
2724 break;
2725
2726 r = -EFAULT;
2727 if (!copy_from_user(hdr, (void __user *)parms.origin,
2728 parms.length))
2729 r = kvm_s390_pv_set_sec_parms(kvm, hdr, parms.length,
2730 &cmd->rc, &cmd->rrc);
2731
2732 vfree(hdr);
2733 break;
2734 }
2735 case KVM_PV_UNPACK: {
2736 struct kvm_s390_pv_unp unp = {};
2737
2738 r = -EINVAL;
2739 if (!kvm_s390_pv_is_protected(kvm) || !mm_is_protected(kvm->mm))
2740 break;
2741
2742 r = -EFAULT;
2743 if (copy_from_user(&unp, argp, sizeof(unp)))
2744 break;
2745
2746 r = kvm_s390_pv_unpack(kvm, unp.addr, unp.size, unp.tweak,
2747 &cmd->rc, &cmd->rrc);
2748 break;
2749 }
2750 case KVM_PV_VERIFY: {
2751 r = -EINVAL;
2752 if (!kvm_s390_pv_is_protected(kvm))
2753 break;
2754
2755 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2756 UVC_CMD_VERIFY_IMG, &cmd->rc, &cmd->rrc);
2757 KVM_UV_EVENT(kvm, 3, "PROTVIRT VERIFY: rc %x rrc %x", cmd->rc,
2758 cmd->rrc);
2759 break;
2760 }
2761 case KVM_PV_PREP_RESET: {
2762 r = -EINVAL;
2763 if (!kvm_s390_pv_is_protected(kvm))
2764 break;
2765
2766 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2767 UVC_CMD_PREPARE_RESET, &cmd->rc, &cmd->rrc);
2768 KVM_UV_EVENT(kvm, 3, "PROTVIRT PREP RESET: rc %x rrc %x",
2769 cmd->rc, cmd->rrc);
2770 break;
2771 }
2772 case KVM_PV_UNSHARE_ALL: {
2773 r = -EINVAL;
2774 if (!kvm_s390_pv_is_protected(kvm))
2775 break;
2776
2777 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2778 UVC_CMD_SET_UNSHARE_ALL, &cmd->rc, &cmd->rrc);
2779 KVM_UV_EVENT(kvm, 3, "PROTVIRT UNSHARE: rc %x rrc %x",
2780 cmd->rc, cmd->rrc);
2781 break;
2782 }
2783 case KVM_PV_INFO: {
2784 struct kvm_s390_pv_info info = {};
2785 ssize_t data_len;
2786
2787 /*
2788 * No need to check the VM protection here.
2789 *
2790 * Maybe user space wants to query some of the data
2791 * when the VM is still unprotected. If we see the
2792 * need to fence a new data command we can still
2793 * return an error in the info handler.
2794 */
2795
2796 r = -EFAULT;
2797 if (copy_from_user(&info, argp, sizeof(info.header)))
2798 break;
2799
2800 r = -EINVAL;
2801 if (info.header.len_max < sizeof(info.header))
2802 break;
2803
2804 data_len = kvm_s390_handle_pv_info(&info);
2805 if (data_len < 0) {
2806 r = data_len;
2807 break;
2808 }
2809 /*
2810 * If a data command struct is extended (multiple
2811 * times) this can be used to determine how much of it
2812 * is valid.
2813 */
2814 info.header.len_written = data_len;
2815
2816 r = -EFAULT;
2817 if (copy_to_user(argp, &info, data_len))
2818 break;
2819
2820 r = 0;
2821 break;
2822 }
2823 case KVM_PV_DUMP: {
2824 struct kvm_s390_pv_dmp dmp;
2825
2826 r = -EINVAL;
2827 if (!kvm_s390_pv_is_protected(kvm))
2828 break;
2829
2830 r = -EFAULT;
2831 if (copy_from_user(&dmp, argp, sizeof(dmp)))
2832 break;
2833
2834 r = kvm_s390_pv_dmp(kvm, cmd, dmp);
2835 if (r)
2836 break;
2837
2838 if (copy_to_user(argp, &dmp, sizeof(dmp))) {
2839 r = -EFAULT;
2840 break;
2841 }
2842
2843 break;
2844 }
2845 default:
2846 r = -ENOTTY;
2847 }
2848 if (need_lock)
2849 mutex_unlock(&kvm->lock);
2850
2851 return r;
2852 }
2853
mem_op_validate_common(struct kvm_s390_mem_op * mop,u64 supported_flags)2854 static int mem_op_validate_common(struct kvm_s390_mem_op *mop, u64 supported_flags)
2855 {
2856 if (mop->flags & ~supported_flags || !mop->size)
2857 return -EINVAL;
2858 if (mop->size > MEM_OP_MAX_SIZE)
2859 return -E2BIG;
2860 if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) {
2861 if (mop->key > 0xf)
2862 return -EINVAL;
2863 } else {
2864 mop->key = 0;
2865 }
2866 return 0;
2867 }
2868
kvm_s390_vm_mem_op_abs(struct kvm * kvm,struct kvm_s390_mem_op * mop)2869 static int kvm_s390_vm_mem_op_abs(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2870 {
2871 void __user *uaddr = (void __user *)mop->buf;
2872 enum gacc_mode acc_mode;
2873 void *tmpbuf = NULL;
2874 int r, srcu_idx;
2875
2876 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION |
2877 KVM_S390_MEMOP_F_CHECK_ONLY);
2878 if (r)
2879 return r;
2880
2881 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2882 tmpbuf = vmalloc(mop->size);
2883 if (!tmpbuf)
2884 return -ENOMEM;
2885 }
2886
2887 srcu_idx = srcu_read_lock(&kvm->srcu);
2888
2889 if (!kvm_is_gpa_in_memslot(kvm, mop->gaddr)) {
2890 r = PGM_ADDRESSING;
2891 goto out_unlock;
2892 }
2893
2894 acc_mode = mop->op == KVM_S390_MEMOP_ABSOLUTE_READ ? GACC_FETCH : GACC_STORE;
2895 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2896 r = check_gpa_range(kvm, mop->gaddr, mop->size, acc_mode, mop->key);
2897 goto out_unlock;
2898 }
2899 if (acc_mode == GACC_FETCH) {
2900 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2901 mop->size, GACC_FETCH, mop->key);
2902 if (r)
2903 goto out_unlock;
2904 if (copy_to_user(uaddr, tmpbuf, mop->size))
2905 r = -EFAULT;
2906 } else {
2907 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2908 r = -EFAULT;
2909 goto out_unlock;
2910 }
2911 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2912 mop->size, GACC_STORE, mop->key);
2913 }
2914
2915 out_unlock:
2916 srcu_read_unlock(&kvm->srcu, srcu_idx);
2917
2918 vfree(tmpbuf);
2919 return r;
2920 }
2921
kvm_s390_vm_mem_op_cmpxchg(struct kvm * kvm,struct kvm_s390_mem_op * mop)2922 static int kvm_s390_vm_mem_op_cmpxchg(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2923 {
2924 void __user *uaddr = (void __user *)mop->buf;
2925 void __user *old_addr = (void __user *)mop->old_addr;
2926 union {
2927 __uint128_t quad;
2928 char raw[sizeof(__uint128_t)];
2929 } old = { .quad = 0}, new = { .quad = 0 };
2930 unsigned int off_in_quad = sizeof(new) - mop->size;
2931 int r, srcu_idx;
2932 bool success;
2933
2934 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION);
2935 if (r)
2936 return r;
2937 /*
2938 * This validates off_in_quad. Checking that size is a power
2939 * of two is not necessary, as cmpxchg_guest_abs_with_key
2940 * takes care of that
2941 */
2942 if (mop->size > sizeof(new))
2943 return -EINVAL;
2944 if (copy_from_user(&new.raw[off_in_quad], uaddr, mop->size))
2945 return -EFAULT;
2946 if (copy_from_user(&old.raw[off_in_quad], old_addr, mop->size))
2947 return -EFAULT;
2948
2949 srcu_idx = srcu_read_lock(&kvm->srcu);
2950
2951 if (!kvm_is_gpa_in_memslot(kvm, mop->gaddr)) {
2952 r = PGM_ADDRESSING;
2953 goto out_unlock;
2954 }
2955
2956 r = cmpxchg_guest_abs_with_key(kvm, mop->gaddr, mop->size, &old.quad,
2957 new.quad, mop->key, &success);
2958 if (!success && copy_to_user(old_addr, &old.raw[off_in_quad], mop->size))
2959 r = -EFAULT;
2960
2961 out_unlock:
2962 srcu_read_unlock(&kvm->srcu, srcu_idx);
2963 return r;
2964 }
2965
kvm_s390_vm_mem_op(struct kvm * kvm,struct kvm_s390_mem_op * mop)2966 static int kvm_s390_vm_mem_op(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2967 {
2968 /*
2969 * This is technically a heuristic only, if the kvm->lock is not
2970 * taken, it is not guaranteed that the vm is/remains non-protected.
2971 * This is ok from a kernel perspective, wrongdoing is detected
2972 * on the access, -EFAULT is returned and the vm may crash the
2973 * next time it accesses the memory in question.
2974 * There is no sane usecase to do switching and a memop on two
2975 * different CPUs at the same time.
2976 */
2977 if (kvm_s390_pv_get_handle(kvm))
2978 return -EINVAL;
2979
2980 switch (mop->op) {
2981 case KVM_S390_MEMOP_ABSOLUTE_READ:
2982 case KVM_S390_MEMOP_ABSOLUTE_WRITE:
2983 return kvm_s390_vm_mem_op_abs(kvm, mop);
2984 case KVM_S390_MEMOP_ABSOLUTE_CMPXCHG:
2985 return kvm_s390_vm_mem_op_cmpxchg(kvm, mop);
2986 default:
2987 return -EINVAL;
2988 }
2989 }
2990
kvm_arch_vm_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2991 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
2992 {
2993 struct kvm *kvm = filp->private_data;
2994 void __user *argp = (void __user *)arg;
2995 struct kvm_device_attr attr;
2996 int r;
2997
2998 switch (ioctl) {
2999 case KVM_S390_INTERRUPT: {
3000 struct kvm_s390_interrupt s390int;
3001
3002 r = -EFAULT;
3003 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3004 break;
3005 r = kvm_s390_inject_vm(kvm, &s390int);
3006 break;
3007 }
3008 case KVM_CREATE_IRQCHIP: {
3009 r = -EINVAL;
3010 if (kvm->arch.use_irqchip)
3011 r = 0;
3012 break;
3013 }
3014 case KVM_SET_DEVICE_ATTR: {
3015 r = -EFAULT;
3016 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3017 break;
3018 r = kvm_s390_vm_set_attr(kvm, &attr);
3019 break;
3020 }
3021 case KVM_GET_DEVICE_ATTR: {
3022 r = -EFAULT;
3023 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3024 break;
3025 r = kvm_s390_vm_get_attr(kvm, &attr);
3026 break;
3027 }
3028 case KVM_HAS_DEVICE_ATTR: {
3029 r = -EFAULT;
3030 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3031 break;
3032 r = kvm_s390_vm_has_attr(kvm, &attr);
3033 break;
3034 }
3035 case KVM_S390_GET_SKEYS: {
3036 struct kvm_s390_skeys args;
3037
3038 r = -EFAULT;
3039 if (copy_from_user(&args, argp,
3040 sizeof(struct kvm_s390_skeys)))
3041 break;
3042 r = kvm_s390_get_skeys(kvm, &args);
3043 break;
3044 }
3045 case KVM_S390_SET_SKEYS: {
3046 struct kvm_s390_skeys args;
3047
3048 r = -EFAULT;
3049 if (copy_from_user(&args, argp,
3050 sizeof(struct kvm_s390_skeys)))
3051 break;
3052 r = kvm_s390_set_skeys(kvm, &args);
3053 break;
3054 }
3055 case KVM_S390_GET_CMMA_BITS: {
3056 struct kvm_s390_cmma_log args;
3057
3058 r = -EFAULT;
3059 if (copy_from_user(&args, argp, sizeof(args)))
3060 break;
3061 mutex_lock(&kvm->slots_lock);
3062 r = kvm_s390_get_cmma_bits(kvm, &args);
3063 mutex_unlock(&kvm->slots_lock);
3064 if (!r) {
3065 r = copy_to_user(argp, &args, sizeof(args));
3066 if (r)
3067 r = -EFAULT;
3068 }
3069 break;
3070 }
3071 case KVM_S390_SET_CMMA_BITS: {
3072 struct kvm_s390_cmma_log args;
3073
3074 r = -EFAULT;
3075 if (copy_from_user(&args, argp, sizeof(args)))
3076 break;
3077 mutex_lock(&kvm->slots_lock);
3078 r = kvm_s390_set_cmma_bits(kvm, &args);
3079 mutex_unlock(&kvm->slots_lock);
3080 break;
3081 }
3082 case KVM_S390_PV_COMMAND: {
3083 struct kvm_pv_cmd args;
3084
3085 /* protvirt means user cpu state */
3086 kvm_s390_set_user_cpu_state_ctrl(kvm);
3087 r = 0;
3088 if (!is_prot_virt_host()) {
3089 r = -EINVAL;
3090 break;
3091 }
3092 if (copy_from_user(&args, argp, sizeof(args))) {
3093 r = -EFAULT;
3094 break;
3095 }
3096 if (args.flags) {
3097 r = -EINVAL;
3098 break;
3099 }
3100 /* must be called without kvm->lock */
3101 r = kvm_s390_handle_pv(kvm, &args);
3102 if (copy_to_user(argp, &args, sizeof(args))) {
3103 r = -EFAULT;
3104 break;
3105 }
3106 break;
3107 }
3108 case KVM_S390_MEM_OP: {
3109 struct kvm_s390_mem_op mem_op;
3110
3111 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3112 r = kvm_s390_vm_mem_op(kvm, &mem_op);
3113 else
3114 r = -EFAULT;
3115 break;
3116 }
3117 case KVM_S390_ZPCI_OP: {
3118 struct kvm_s390_zpci_op args;
3119
3120 r = -EINVAL;
3121 if (!IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3122 break;
3123 if (copy_from_user(&args, argp, sizeof(args))) {
3124 r = -EFAULT;
3125 break;
3126 }
3127 r = kvm_s390_pci_zpci_op(kvm, &args);
3128 break;
3129 }
3130 default:
3131 r = -ENOTTY;
3132 }
3133
3134 return r;
3135 }
3136
kvm_s390_apxa_installed(void)3137 static int kvm_s390_apxa_installed(void)
3138 {
3139 struct ap_config_info info;
3140
3141 if (ap_instructions_available()) {
3142 if (ap_qci(&info) == 0)
3143 return info.apxa;
3144 }
3145
3146 return 0;
3147 }
3148
3149 /*
3150 * The format of the crypto control block (CRYCB) is specified in the 3 low
3151 * order bits of the CRYCB designation (CRYCBD) field as follows:
3152 * Format 0: Neither the message security assist extension 3 (MSAX3) nor the
3153 * AP extended addressing (APXA) facility are installed.
3154 * Format 1: The APXA facility is not installed but the MSAX3 facility is.
3155 * Format 2: Both the APXA and MSAX3 facilities are installed
3156 */
kvm_s390_set_crycb_format(struct kvm * kvm)3157 static void kvm_s390_set_crycb_format(struct kvm *kvm)
3158 {
3159 kvm->arch.crypto.crycbd = virt_to_phys(kvm->arch.crypto.crycb);
3160
3161 /* Clear the CRYCB format bits - i.e., set format 0 by default */
3162 kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK);
3163
3164 /* Check whether MSAX3 is installed */
3165 if (!test_kvm_facility(kvm, 76))
3166 return;
3167
3168 if (kvm_s390_apxa_installed())
3169 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
3170 else
3171 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
3172 }
3173
3174 /*
3175 * kvm_arch_crypto_set_masks
3176 *
3177 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3178 * to be set.
3179 * @apm: the mask identifying the accessible AP adapters
3180 * @aqm: the mask identifying the accessible AP domains
3181 * @adm: the mask identifying the accessible AP control domains
3182 *
3183 * Set the masks that identify the adapters, domains and control domains to
3184 * which the KVM guest is granted access.
3185 *
3186 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3187 * function.
3188 */
kvm_arch_crypto_set_masks(struct kvm * kvm,unsigned long * apm,unsigned long * aqm,unsigned long * adm)3189 void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm,
3190 unsigned long *aqm, unsigned long *adm)
3191 {
3192 struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb;
3193
3194 kvm_s390_vcpu_block_all(kvm);
3195
3196 switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) {
3197 case CRYCB_FORMAT2: /* APCB1 use 256 bits */
3198 memcpy(crycb->apcb1.apm, apm, 32);
3199 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx",
3200 apm[0], apm[1], apm[2], apm[3]);
3201 memcpy(crycb->apcb1.aqm, aqm, 32);
3202 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx",
3203 aqm[0], aqm[1], aqm[2], aqm[3]);
3204 memcpy(crycb->apcb1.adm, adm, 32);
3205 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx",
3206 adm[0], adm[1], adm[2], adm[3]);
3207 break;
3208 case CRYCB_FORMAT1:
3209 case CRYCB_FORMAT0: /* Fall through both use APCB0 */
3210 memcpy(crycb->apcb0.apm, apm, 8);
3211 memcpy(crycb->apcb0.aqm, aqm, 2);
3212 memcpy(crycb->apcb0.adm, adm, 2);
3213 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x",
3214 apm[0], *((unsigned short *)aqm),
3215 *((unsigned short *)adm));
3216 break;
3217 default: /* Can not happen */
3218 break;
3219 }
3220
3221 /* recreate the shadow crycb for each vcpu */
3222 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3223 kvm_s390_vcpu_unblock_all(kvm);
3224 }
3225 EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks);
3226
3227 /*
3228 * kvm_arch_crypto_clear_masks
3229 *
3230 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3231 * to be cleared.
3232 *
3233 * Clear the masks that identify the adapters, domains and control domains to
3234 * which the KVM guest is granted access.
3235 *
3236 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3237 * function.
3238 */
kvm_arch_crypto_clear_masks(struct kvm * kvm)3239 void kvm_arch_crypto_clear_masks(struct kvm *kvm)
3240 {
3241 kvm_s390_vcpu_block_all(kvm);
3242
3243 memset(&kvm->arch.crypto.crycb->apcb0, 0,
3244 sizeof(kvm->arch.crypto.crycb->apcb0));
3245 memset(&kvm->arch.crypto.crycb->apcb1, 0,
3246 sizeof(kvm->arch.crypto.crycb->apcb1));
3247
3248 VM_EVENT(kvm, 3, "%s", "CLR CRYCB:");
3249 /* recreate the shadow crycb for each vcpu */
3250 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3251 kvm_s390_vcpu_unblock_all(kvm);
3252 }
3253 EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks);
3254
kvm_s390_get_initial_cpuid(void)3255 static u64 kvm_s390_get_initial_cpuid(void)
3256 {
3257 struct cpuid cpuid;
3258
3259 get_cpu_id(&cpuid);
3260 cpuid.version = 0xff;
3261 return *((u64 *) &cpuid);
3262 }
3263
kvm_s390_crypto_init(struct kvm * kvm)3264 static void kvm_s390_crypto_init(struct kvm *kvm)
3265 {
3266 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
3267 kvm_s390_set_crycb_format(kvm);
3268 init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
3269
3270 if (!test_kvm_facility(kvm, 76))
3271 return;
3272
3273 /* Enable AES/DEA protected key functions by default */
3274 kvm->arch.crypto.aes_kw = 1;
3275 kvm->arch.crypto.dea_kw = 1;
3276 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
3277 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
3278 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
3279 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
3280 }
3281
sca_dispose(struct kvm * kvm)3282 static void sca_dispose(struct kvm *kvm)
3283 {
3284 if (kvm->arch.use_esca)
3285 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
3286 else
3287 free_page((unsigned long)(kvm->arch.sca));
3288 kvm->arch.sca = NULL;
3289 }
3290
kvm_arch_free_vm(struct kvm * kvm)3291 void kvm_arch_free_vm(struct kvm *kvm)
3292 {
3293 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3294 kvm_s390_pci_clear_list(kvm);
3295
3296 __kvm_arch_free_vm(kvm);
3297 }
3298
kvm_arch_init_vm(struct kvm * kvm,unsigned long type)3299 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
3300 {
3301 gfp_t alloc_flags = GFP_KERNEL_ACCOUNT;
3302 int i, rc;
3303 char debug_name[16];
3304 static unsigned long sca_offset;
3305
3306 rc = -EINVAL;
3307 #ifdef CONFIG_KVM_S390_UCONTROL
3308 if (type & ~KVM_VM_S390_UCONTROL)
3309 goto out_err;
3310 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
3311 goto out_err;
3312 #else
3313 if (type)
3314 goto out_err;
3315 #endif
3316
3317 rc = s390_enable_sie();
3318 if (rc)
3319 goto out_err;
3320
3321 rc = -ENOMEM;
3322
3323 if (!sclp.has_64bscao)
3324 alloc_flags |= GFP_DMA;
3325 rwlock_init(&kvm->arch.sca_lock);
3326 /* start with basic SCA */
3327 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
3328 if (!kvm->arch.sca)
3329 goto out_err;
3330 mutex_lock(&kvm_lock);
3331 sca_offset += 16;
3332 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
3333 sca_offset = 0;
3334 kvm->arch.sca = (struct bsca_block *)
3335 ((char *) kvm->arch.sca + sca_offset);
3336 mutex_unlock(&kvm_lock);
3337
3338 sprintf(debug_name, "kvm-%u", current->pid);
3339
3340 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
3341 if (!kvm->arch.dbf)
3342 goto out_err;
3343
3344 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
3345 kvm->arch.sie_page2 =
3346 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3347 if (!kvm->arch.sie_page2)
3348 goto out_err;
3349
3350 kvm->arch.sie_page2->kvm = kvm;
3351 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
3352
3353 for (i = 0; i < kvm_s390_fac_size(); i++) {
3354 kvm->arch.model.fac_mask[i] = stfle_fac_list[i] &
3355 (kvm_s390_fac_base[i] |
3356 kvm_s390_fac_ext[i]);
3357 kvm->arch.model.fac_list[i] = stfle_fac_list[i] &
3358 kvm_s390_fac_base[i];
3359 }
3360 kvm->arch.model.subfuncs = kvm_s390_available_subfunc;
3361
3362 /* we are always in czam mode - even on pre z14 machines */
3363 set_kvm_facility(kvm->arch.model.fac_mask, 138);
3364 set_kvm_facility(kvm->arch.model.fac_list, 138);
3365 /* we emulate STHYI in kvm */
3366 set_kvm_facility(kvm->arch.model.fac_mask, 74);
3367 set_kvm_facility(kvm->arch.model.fac_list, 74);
3368 if (MACHINE_HAS_TLB_GUEST) {
3369 set_kvm_facility(kvm->arch.model.fac_mask, 147);
3370 set_kvm_facility(kvm->arch.model.fac_list, 147);
3371 }
3372
3373 if (css_general_characteristics.aiv && test_facility(65))
3374 set_kvm_facility(kvm->arch.model.fac_mask, 65);
3375
3376 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
3377 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
3378
3379 kvm->arch.model.uv_feat_guest.feat = 0;
3380
3381 kvm_s390_crypto_init(kvm);
3382
3383 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3384 mutex_lock(&kvm->lock);
3385 kvm_s390_pci_init_list(kvm);
3386 kvm_s390_vcpu_pci_enable_interp(kvm);
3387 mutex_unlock(&kvm->lock);
3388 }
3389
3390 mutex_init(&kvm->arch.float_int.ais_lock);
3391 spin_lock_init(&kvm->arch.float_int.lock);
3392 for (i = 0; i < FIRQ_LIST_COUNT; i++)
3393 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
3394 init_waitqueue_head(&kvm->arch.ipte_wq);
3395 mutex_init(&kvm->arch.ipte_mutex);
3396
3397 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
3398 VM_EVENT(kvm, 3, "vm created with type %lu", type);
3399
3400 if (type & KVM_VM_S390_UCONTROL) {
3401 kvm->arch.gmap = NULL;
3402 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
3403 } else {
3404 if (sclp.hamax == U64_MAX)
3405 kvm->arch.mem_limit = TASK_SIZE_MAX;
3406 else
3407 kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
3408 sclp.hamax + 1);
3409 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
3410 if (!kvm->arch.gmap)
3411 goto out_err;
3412 kvm->arch.gmap->private = kvm;
3413 kvm->arch.gmap->pfault_enabled = 0;
3414 }
3415
3416 kvm->arch.use_pfmfi = sclp.has_pfmfi;
3417 kvm->arch.use_skf = sclp.has_skey;
3418 spin_lock_init(&kvm->arch.start_stop_lock);
3419 kvm_s390_vsie_init(kvm);
3420 if (use_gisa)
3421 kvm_s390_gisa_init(kvm);
3422 INIT_LIST_HEAD(&kvm->arch.pv.need_cleanup);
3423 kvm->arch.pv.set_aside = NULL;
3424 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
3425
3426 return 0;
3427 out_err:
3428 free_page((unsigned long)kvm->arch.sie_page2);
3429 debug_unregister(kvm->arch.dbf);
3430 sca_dispose(kvm);
3431 KVM_EVENT(3, "creation of vm failed: %d", rc);
3432 return rc;
3433 }
3434
kvm_arch_vcpu_destroy(struct kvm_vcpu * vcpu)3435 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3436 {
3437 u16 rc, rrc;
3438
3439 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
3440 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
3441 kvm_s390_clear_local_irqs(vcpu);
3442 kvm_clear_async_pf_completion_queue(vcpu);
3443 if (!kvm_is_ucontrol(vcpu->kvm))
3444 sca_del_vcpu(vcpu);
3445 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3446
3447 if (kvm_is_ucontrol(vcpu->kvm))
3448 gmap_remove(vcpu->arch.gmap);
3449
3450 if (vcpu->kvm->arch.use_cmma)
3451 kvm_s390_vcpu_unsetup_cmma(vcpu);
3452 /* We can not hold the vcpu mutex here, we are already dying */
3453 if (kvm_s390_pv_cpu_get_handle(vcpu))
3454 kvm_s390_pv_destroy_cpu(vcpu, &rc, &rrc);
3455 free_page((unsigned long)(vcpu->arch.sie_block));
3456 }
3457
kvm_arch_destroy_vm(struct kvm * kvm)3458 void kvm_arch_destroy_vm(struct kvm *kvm)
3459 {
3460 u16 rc, rrc;
3461
3462 kvm_destroy_vcpus(kvm);
3463 sca_dispose(kvm);
3464 kvm_s390_gisa_destroy(kvm);
3465 /*
3466 * We are already at the end of life and kvm->lock is not taken.
3467 * This is ok as the file descriptor is closed by now and nobody
3468 * can mess with the pv state.
3469 */
3470 kvm_s390_pv_deinit_cleanup_all(kvm, &rc, &rrc);
3471 /*
3472 * Remove the mmu notifier only when the whole KVM VM is torn down,
3473 * and only if one was registered to begin with. If the VM is
3474 * currently not protected, but has been previously been protected,
3475 * then it's possible that the notifier is still registered.
3476 */
3477 if (kvm->arch.pv.mmu_notifier.ops)
3478 mmu_notifier_unregister(&kvm->arch.pv.mmu_notifier, kvm->mm);
3479
3480 debug_unregister(kvm->arch.dbf);
3481 free_page((unsigned long)kvm->arch.sie_page2);
3482 if (!kvm_is_ucontrol(kvm))
3483 gmap_remove(kvm->arch.gmap);
3484 kvm_s390_destroy_adapters(kvm);
3485 kvm_s390_clear_float_irqs(kvm);
3486 kvm_s390_vsie_destroy(kvm);
3487 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
3488 }
3489
3490 /* Section: vcpu related */
__kvm_ucontrol_vcpu_init(struct kvm_vcpu * vcpu)3491 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
3492 {
3493 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
3494 if (!vcpu->arch.gmap)
3495 return -ENOMEM;
3496 vcpu->arch.gmap->private = vcpu->kvm;
3497
3498 return 0;
3499 }
3500
sca_del_vcpu(struct kvm_vcpu * vcpu)3501 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
3502 {
3503 if (!kvm_s390_use_sca_entries())
3504 return;
3505 read_lock(&vcpu->kvm->arch.sca_lock);
3506 if (vcpu->kvm->arch.use_esca) {
3507 struct esca_block *sca = vcpu->kvm->arch.sca;
3508
3509 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3510 sca->cpu[vcpu->vcpu_id].sda = 0;
3511 } else {
3512 struct bsca_block *sca = vcpu->kvm->arch.sca;
3513
3514 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3515 sca->cpu[vcpu->vcpu_id].sda = 0;
3516 }
3517 read_unlock(&vcpu->kvm->arch.sca_lock);
3518 }
3519
sca_add_vcpu(struct kvm_vcpu * vcpu)3520 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
3521 {
3522 if (!kvm_s390_use_sca_entries()) {
3523 phys_addr_t sca_phys = virt_to_phys(vcpu->kvm->arch.sca);
3524
3525 /* we still need the basic sca for the ipte control */
3526 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3527 vcpu->arch.sie_block->scaol = sca_phys;
3528 return;
3529 }
3530 read_lock(&vcpu->kvm->arch.sca_lock);
3531 if (vcpu->kvm->arch.use_esca) {
3532 struct esca_block *sca = vcpu->kvm->arch.sca;
3533 phys_addr_t sca_phys = virt_to_phys(sca);
3534
3535 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3536 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3537 vcpu->arch.sie_block->scaol = sca_phys & ESCA_SCAOL_MASK;
3538 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3539 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3540 } else {
3541 struct bsca_block *sca = vcpu->kvm->arch.sca;
3542 phys_addr_t sca_phys = virt_to_phys(sca);
3543
3544 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3545 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3546 vcpu->arch.sie_block->scaol = sca_phys;
3547 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3548 }
3549 read_unlock(&vcpu->kvm->arch.sca_lock);
3550 }
3551
3552 /* Basic SCA to Extended SCA data copy routines */
sca_copy_entry(struct esca_entry * d,struct bsca_entry * s)3553 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
3554 {
3555 d->sda = s->sda;
3556 d->sigp_ctrl.c = s->sigp_ctrl.c;
3557 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
3558 }
3559
sca_copy_b_to_e(struct esca_block * d,struct bsca_block * s)3560 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
3561 {
3562 int i;
3563
3564 d->ipte_control = s->ipte_control;
3565 d->mcn[0] = s->mcn;
3566 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
3567 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
3568 }
3569
sca_switch_to_extended(struct kvm * kvm)3570 static int sca_switch_to_extended(struct kvm *kvm)
3571 {
3572 struct bsca_block *old_sca = kvm->arch.sca;
3573 struct esca_block *new_sca;
3574 struct kvm_vcpu *vcpu;
3575 unsigned long vcpu_idx;
3576 u32 scaol, scaoh;
3577 phys_addr_t new_sca_phys;
3578
3579 if (kvm->arch.use_esca)
3580 return 0;
3581
3582 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3583 if (!new_sca)
3584 return -ENOMEM;
3585
3586 new_sca_phys = virt_to_phys(new_sca);
3587 scaoh = new_sca_phys >> 32;
3588 scaol = new_sca_phys & ESCA_SCAOL_MASK;
3589
3590 kvm_s390_vcpu_block_all(kvm);
3591 write_lock(&kvm->arch.sca_lock);
3592
3593 sca_copy_b_to_e(new_sca, old_sca);
3594
3595 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
3596 vcpu->arch.sie_block->scaoh = scaoh;
3597 vcpu->arch.sie_block->scaol = scaol;
3598 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3599 }
3600 kvm->arch.sca = new_sca;
3601 kvm->arch.use_esca = 1;
3602
3603 write_unlock(&kvm->arch.sca_lock);
3604 kvm_s390_vcpu_unblock_all(kvm);
3605
3606 free_page((unsigned long)old_sca);
3607
3608 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
3609 old_sca, kvm->arch.sca);
3610 return 0;
3611 }
3612
sca_can_add_vcpu(struct kvm * kvm,unsigned int id)3613 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
3614 {
3615 int rc;
3616
3617 if (!kvm_s390_use_sca_entries()) {
3618 if (id < KVM_MAX_VCPUS)
3619 return true;
3620 return false;
3621 }
3622 if (id < KVM_S390_BSCA_CPU_SLOTS)
3623 return true;
3624 if (!sclp.has_esca || !sclp.has_64bscao)
3625 return false;
3626
3627 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
3628
3629 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
3630 }
3631
3632 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
__start_cpu_timer_accounting(struct kvm_vcpu * vcpu)3633 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3634 {
3635 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
3636 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3637 vcpu->arch.cputm_start = get_tod_clock_fast();
3638 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3639 }
3640
3641 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
__stop_cpu_timer_accounting(struct kvm_vcpu * vcpu)3642 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3643 {
3644 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
3645 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3646 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3647 vcpu->arch.cputm_start = 0;
3648 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3649 }
3650
3651 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
__enable_cpu_timer_accounting(struct kvm_vcpu * vcpu)3652 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3653 {
3654 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
3655 vcpu->arch.cputm_enabled = true;
3656 __start_cpu_timer_accounting(vcpu);
3657 }
3658
3659 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
__disable_cpu_timer_accounting(struct kvm_vcpu * vcpu)3660 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3661 {
3662 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
3663 __stop_cpu_timer_accounting(vcpu);
3664 vcpu->arch.cputm_enabled = false;
3665 }
3666
enable_cpu_timer_accounting(struct kvm_vcpu * vcpu)3667 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3668 {
3669 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3670 __enable_cpu_timer_accounting(vcpu);
3671 preempt_enable();
3672 }
3673
disable_cpu_timer_accounting(struct kvm_vcpu * vcpu)3674 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3675 {
3676 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3677 __disable_cpu_timer_accounting(vcpu);
3678 preempt_enable();
3679 }
3680
3681 /* set the cpu timer - may only be called from the VCPU thread itself */
kvm_s390_set_cpu_timer(struct kvm_vcpu * vcpu,__u64 cputm)3682 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
3683 {
3684 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3685 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3686 if (vcpu->arch.cputm_enabled)
3687 vcpu->arch.cputm_start = get_tod_clock_fast();
3688 vcpu->arch.sie_block->cputm = cputm;
3689 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3690 preempt_enable();
3691 }
3692
3693 /* update and get the cpu timer - can also be called from other VCPU threads */
kvm_s390_get_cpu_timer(struct kvm_vcpu * vcpu)3694 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
3695 {
3696 unsigned int seq;
3697 __u64 value;
3698
3699 if (unlikely(!vcpu->arch.cputm_enabled))
3700 return vcpu->arch.sie_block->cputm;
3701
3702 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3703 do {
3704 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
3705 /*
3706 * If the writer would ever execute a read in the critical
3707 * section, e.g. in irq context, we have a deadlock.
3708 */
3709 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
3710 value = vcpu->arch.sie_block->cputm;
3711 /* if cputm_start is 0, accounting is being started/stopped */
3712 if (likely(vcpu->arch.cputm_start))
3713 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3714 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
3715 preempt_enable();
3716 return value;
3717 }
3718
kvm_arch_vcpu_load(struct kvm_vcpu * vcpu,int cpu)3719 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3720 {
3721
3722 gmap_enable(vcpu->arch.enabled_gmap);
3723 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
3724 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3725 __start_cpu_timer_accounting(vcpu);
3726 vcpu->cpu = cpu;
3727 }
3728
kvm_arch_vcpu_put(struct kvm_vcpu * vcpu)3729 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3730 {
3731 vcpu->cpu = -1;
3732 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3733 __stop_cpu_timer_accounting(vcpu);
3734 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
3735 vcpu->arch.enabled_gmap = gmap_get_enabled();
3736 gmap_disable(vcpu->arch.enabled_gmap);
3737
3738 }
3739
kvm_arch_vcpu_postcreate(struct kvm_vcpu * vcpu)3740 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
3741 {
3742 mutex_lock(&vcpu->kvm->lock);
3743 preempt_disable();
3744 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
3745 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
3746 preempt_enable();
3747 mutex_unlock(&vcpu->kvm->lock);
3748 if (!kvm_is_ucontrol(vcpu->kvm)) {
3749 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
3750 sca_add_vcpu(vcpu);
3751 }
3752 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
3753 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3754 /* make vcpu_load load the right gmap on the first trigger */
3755 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
3756 }
3757
kvm_has_pckmo_subfunc(struct kvm * kvm,unsigned long nr)3758 static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr)
3759 {
3760 if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) &&
3761 test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo))
3762 return true;
3763 return false;
3764 }
3765
kvm_has_pckmo_ecc(struct kvm * kvm)3766 static bool kvm_has_pckmo_ecc(struct kvm *kvm)
3767 {
3768 /* At least one ECC subfunction must be present */
3769 return kvm_has_pckmo_subfunc(kvm, 32) ||
3770 kvm_has_pckmo_subfunc(kvm, 33) ||
3771 kvm_has_pckmo_subfunc(kvm, 34) ||
3772 kvm_has_pckmo_subfunc(kvm, 40) ||
3773 kvm_has_pckmo_subfunc(kvm, 41);
3774
3775 }
3776
kvm_s390_vcpu_crypto_setup(struct kvm_vcpu * vcpu)3777 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
3778 {
3779 /*
3780 * If the AP instructions are not being interpreted and the MSAX3
3781 * facility is not configured for the guest, there is nothing to set up.
3782 */
3783 if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76))
3784 return;
3785
3786 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
3787 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
3788 vcpu->arch.sie_block->eca &= ~ECA_APIE;
3789 vcpu->arch.sie_block->ecd &= ~ECD_ECC;
3790
3791 if (vcpu->kvm->arch.crypto.apie)
3792 vcpu->arch.sie_block->eca |= ECA_APIE;
3793
3794 /* Set up protected key support */
3795 if (vcpu->kvm->arch.crypto.aes_kw) {
3796 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
3797 /* ecc is also wrapped with AES key */
3798 if (kvm_has_pckmo_ecc(vcpu->kvm))
3799 vcpu->arch.sie_block->ecd |= ECD_ECC;
3800 }
3801
3802 if (vcpu->kvm->arch.crypto.dea_kw)
3803 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
3804 }
3805
kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu * vcpu)3806 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
3807 {
3808 free_page((unsigned long)phys_to_virt(vcpu->arch.sie_block->cbrlo));
3809 vcpu->arch.sie_block->cbrlo = 0;
3810 }
3811
kvm_s390_vcpu_setup_cmma(struct kvm_vcpu * vcpu)3812 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
3813 {
3814 void *cbrlo_page = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
3815
3816 if (!cbrlo_page)
3817 return -ENOMEM;
3818
3819 vcpu->arch.sie_block->cbrlo = virt_to_phys(cbrlo_page);
3820 return 0;
3821 }
3822
kvm_s390_vcpu_setup_model(struct kvm_vcpu * vcpu)3823 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
3824 {
3825 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
3826
3827 vcpu->arch.sie_block->ibc = model->ibc;
3828 if (test_kvm_facility(vcpu->kvm, 7))
3829 vcpu->arch.sie_block->fac = virt_to_phys(model->fac_list);
3830 }
3831
kvm_s390_vcpu_setup(struct kvm_vcpu * vcpu)3832 static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu)
3833 {
3834 int rc = 0;
3835 u16 uvrc, uvrrc;
3836
3837 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
3838 CPUSTAT_SM |
3839 CPUSTAT_STOPPED);
3840
3841 if (test_kvm_facility(vcpu->kvm, 78))
3842 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
3843 else if (test_kvm_facility(vcpu->kvm, 8))
3844 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
3845
3846 kvm_s390_vcpu_setup_model(vcpu);
3847
3848 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
3849 if (MACHINE_HAS_ESOP)
3850 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
3851 if (test_kvm_facility(vcpu->kvm, 9))
3852 vcpu->arch.sie_block->ecb |= ECB_SRSI;
3853 if (test_kvm_facility(vcpu->kvm, 11))
3854 vcpu->arch.sie_block->ecb |= ECB_PTF;
3855 if (test_kvm_facility(vcpu->kvm, 73))
3856 vcpu->arch.sie_block->ecb |= ECB_TE;
3857 if (!kvm_is_ucontrol(vcpu->kvm))
3858 vcpu->arch.sie_block->ecb |= ECB_SPECI;
3859
3860 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
3861 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
3862 if (test_kvm_facility(vcpu->kvm, 130))
3863 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
3864 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
3865 if (sclp.has_cei)
3866 vcpu->arch.sie_block->eca |= ECA_CEI;
3867 if (sclp.has_ib)
3868 vcpu->arch.sie_block->eca |= ECA_IB;
3869 if (sclp.has_siif)
3870 vcpu->arch.sie_block->eca |= ECA_SII;
3871 if (sclp.has_sigpif)
3872 vcpu->arch.sie_block->eca |= ECA_SIGPI;
3873 if (test_kvm_facility(vcpu->kvm, 129)) {
3874 vcpu->arch.sie_block->eca |= ECA_VX;
3875 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3876 }
3877 if (test_kvm_facility(vcpu->kvm, 139))
3878 vcpu->arch.sie_block->ecd |= ECD_MEF;
3879 if (test_kvm_facility(vcpu->kvm, 156))
3880 vcpu->arch.sie_block->ecd |= ECD_ETOKENF;
3881 if (vcpu->arch.sie_block->gd) {
3882 vcpu->arch.sie_block->eca |= ECA_AIV;
3883 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3884 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3885 }
3886 vcpu->arch.sie_block->sdnxo = virt_to_phys(&vcpu->run->s.regs.sdnx) | SDNXC;
3887 vcpu->arch.sie_block->riccbd = virt_to_phys(&vcpu->run->s.regs.riccb);
3888
3889 if (sclp.has_kss)
3890 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
3891 else
3892 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
3893
3894 if (vcpu->kvm->arch.use_cmma) {
3895 rc = kvm_s390_vcpu_setup_cmma(vcpu);
3896 if (rc)
3897 return rc;
3898 }
3899 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3900 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
3901
3902 vcpu->arch.sie_block->hpid = HPID_KVM;
3903
3904 kvm_s390_vcpu_crypto_setup(vcpu);
3905
3906 kvm_s390_vcpu_pci_setup(vcpu);
3907
3908 mutex_lock(&vcpu->kvm->lock);
3909 if (kvm_s390_pv_is_protected(vcpu->kvm)) {
3910 rc = kvm_s390_pv_create_cpu(vcpu, &uvrc, &uvrrc);
3911 if (rc)
3912 kvm_s390_vcpu_unsetup_cmma(vcpu);
3913 }
3914 mutex_unlock(&vcpu->kvm->lock);
3915
3916 return rc;
3917 }
3918
kvm_arch_vcpu_precreate(struct kvm * kvm,unsigned int id)3919 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
3920 {
3921 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
3922 return -EINVAL;
3923 return 0;
3924 }
3925
kvm_arch_vcpu_create(struct kvm_vcpu * vcpu)3926 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
3927 {
3928 struct sie_page *sie_page;
3929 int rc;
3930
3931 BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
3932 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL_ACCOUNT);
3933 if (!sie_page)
3934 return -ENOMEM;
3935
3936 vcpu->arch.sie_block = &sie_page->sie_block;
3937 vcpu->arch.sie_block->itdba = virt_to_phys(&sie_page->itdb);
3938
3939 /* the real guest size will always be smaller than msl */
3940 vcpu->arch.sie_block->mso = 0;
3941 vcpu->arch.sie_block->msl = sclp.hamax;
3942
3943 vcpu->arch.sie_block->icpua = vcpu->vcpu_id;
3944 spin_lock_init(&vcpu->arch.local_int.lock);
3945 vcpu->arch.sie_block->gd = kvm_s390_get_gisa_desc(vcpu->kvm);
3946 seqcount_init(&vcpu->arch.cputm_seqcount);
3947
3948 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
3949 kvm_clear_async_pf_completion_queue(vcpu);
3950 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
3951 KVM_SYNC_GPRS |
3952 KVM_SYNC_ACRS |
3953 KVM_SYNC_CRS |
3954 KVM_SYNC_ARCH0 |
3955 KVM_SYNC_PFAULT |
3956 KVM_SYNC_DIAG318;
3957 vcpu->arch.acrs_loaded = false;
3958 kvm_s390_set_prefix(vcpu, 0);
3959 if (test_kvm_facility(vcpu->kvm, 64))
3960 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
3961 if (test_kvm_facility(vcpu->kvm, 82))
3962 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
3963 if (test_kvm_facility(vcpu->kvm, 133))
3964 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
3965 if (test_kvm_facility(vcpu->kvm, 156))
3966 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN;
3967 /* fprs can be synchronized via vrs, even if the guest has no vx. With
3968 * cpu_has_vx(), (load|store)_fpu_regs() will work with vrs format.
3969 */
3970 if (cpu_has_vx())
3971 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
3972 else
3973 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
3974
3975 if (kvm_is_ucontrol(vcpu->kvm)) {
3976 rc = __kvm_ucontrol_vcpu_init(vcpu);
3977 if (rc)
3978 goto out_free_sie_block;
3979 }
3980
3981 VM_EVENT(vcpu->kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK",
3982 vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3983 trace_kvm_s390_create_vcpu(vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3984
3985 rc = kvm_s390_vcpu_setup(vcpu);
3986 if (rc)
3987 goto out_ucontrol_uninit;
3988
3989 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3990 return 0;
3991
3992 out_ucontrol_uninit:
3993 if (kvm_is_ucontrol(vcpu->kvm))
3994 gmap_remove(vcpu->arch.gmap);
3995 out_free_sie_block:
3996 free_page((unsigned long)(vcpu->arch.sie_block));
3997 return rc;
3998 }
3999
kvm_arch_vcpu_runnable(struct kvm_vcpu * vcpu)4000 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4001 {
4002 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4003 return kvm_s390_vcpu_has_irq(vcpu, 0);
4004 }
4005
kvm_arch_vcpu_in_kernel(struct kvm_vcpu * vcpu)4006 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
4007 {
4008 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
4009 }
4010
kvm_s390_vcpu_block(struct kvm_vcpu * vcpu)4011 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
4012 {
4013 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4014 exit_sie(vcpu);
4015 }
4016
kvm_s390_vcpu_unblock(struct kvm_vcpu * vcpu)4017 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
4018 {
4019 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4020 }
4021
kvm_s390_vcpu_request(struct kvm_vcpu * vcpu)4022 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
4023 {
4024 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4025 exit_sie(vcpu);
4026 }
4027
kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu * vcpu)4028 bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu)
4029 {
4030 return atomic_read(&vcpu->arch.sie_block->prog20) &
4031 (PROG_BLOCK_SIE | PROG_REQUEST);
4032 }
4033
kvm_s390_vcpu_request_handled(struct kvm_vcpu * vcpu)4034 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
4035 {
4036 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4037 }
4038
4039 /*
4040 * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running.
4041 * If the CPU is not running (e.g. waiting as idle) the function will
4042 * return immediately. */
exit_sie(struct kvm_vcpu * vcpu)4043 void exit_sie(struct kvm_vcpu *vcpu)
4044 {
4045 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
4046 kvm_s390_vsie_kick(vcpu);
4047 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
4048 cpu_relax();
4049 }
4050
4051 /* Kick a guest cpu out of SIE to process a request synchronously */
kvm_s390_sync_request(int req,struct kvm_vcpu * vcpu)4052 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
4053 {
4054 __kvm_make_request(req, vcpu);
4055 kvm_s390_vcpu_request(vcpu);
4056 }
4057
kvm_gmap_notifier(struct gmap * gmap,unsigned long start,unsigned long end)4058 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
4059 unsigned long end)
4060 {
4061 struct kvm *kvm = gmap->private;
4062 struct kvm_vcpu *vcpu;
4063 unsigned long prefix;
4064 unsigned long i;
4065
4066 trace_kvm_s390_gmap_notifier(start, end, gmap_is_shadow(gmap));
4067
4068 if (gmap_is_shadow(gmap))
4069 return;
4070 if (start >= 1UL << 31)
4071 /* We are only interested in prefix pages */
4072 return;
4073 kvm_for_each_vcpu(i, vcpu, kvm) {
4074 /* match against both prefix pages */
4075 prefix = kvm_s390_get_prefix(vcpu);
4076 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
4077 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
4078 start, end);
4079 kvm_s390_sync_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4080 }
4081 }
4082 }
4083
kvm_arch_no_poll(struct kvm_vcpu * vcpu)4084 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
4085 {
4086 /* do not poll with more than halt_poll_max_steal percent of steal time */
4087 if (get_lowcore()->avg_steal_timer * 100 / (TICK_USEC << 12) >=
4088 READ_ONCE(halt_poll_max_steal)) {
4089 vcpu->stat.halt_no_poll_steal++;
4090 return true;
4091 }
4092 return false;
4093 }
4094
kvm_arch_vcpu_should_kick(struct kvm_vcpu * vcpu)4095 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
4096 {
4097 /* kvm common code refers to this, but never calls it */
4098 BUG();
4099 return 0;
4100 }
4101
kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu * vcpu,struct kvm_one_reg * reg)4102 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
4103 struct kvm_one_reg *reg)
4104 {
4105 int r = -EINVAL;
4106
4107 switch (reg->id) {
4108 case KVM_REG_S390_TODPR:
4109 r = put_user(vcpu->arch.sie_block->todpr,
4110 (u32 __user *)reg->addr);
4111 break;
4112 case KVM_REG_S390_EPOCHDIFF:
4113 r = put_user(vcpu->arch.sie_block->epoch,
4114 (u64 __user *)reg->addr);
4115 break;
4116 case KVM_REG_S390_CPU_TIMER:
4117 r = put_user(kvm_s390_get_cpu_timer(vcpu),
4118 (u64 __user *)reg->addr);
4119 break;
4120 case KVM_REG_S390_CLOCK_COMP:
4121 r = put_user(vcpu->arch.sie_block->ckc,
4122 (u64 __user *)reg->addr);
4123 break;
4124 case KVM_REG_S390_PFTOKEN:
4125 r = put_user(vcpu->arch.pfault_token,
4126 (u64 __user *)reg->addr);
4127 break;
4128 case KVM_REG_S390_PFCOMPARE:
4129 r = put_user(vcpu->arch.pfault_compare,
4130 (u64 __user *)reg->addr);
4131 break;
4132 case KVM_REG_S390_PFSELECT:
4133 r = put_user(vcpu->arch.pfault_select,
4134 (u64 __user *)reg->addr);
4135 break;
4136 case KVM_REG_S390_PP:
4137 r = put_user(vcpu->arch.sie_block->pp,
4138 (u64 __user *)reg->addr);
4139 break;
4140 case KVM_REG_S390_GBEA:
4141 r = put_user(vcpu->arch.sie_block->gbea,
4142 (u64 __user *)reg->addr);
4143 break;
4144 default:
4145 break;
4146 }
4147
4148 return r;
4149 }
4150
kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu * vcpu,struct kvm_one_reg * reg)4151 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
4152 struct kvm_one_reg *reg)
4153 {
4154 int r = -EINVAL;
4155 __u64 val;
4156
4157 switch (reg->id) {
4158 case KVM_REG_S390_TODPR:
4159 r = get_user(vcpu->arch.sie_block->todpr,
4160 (u32 __user *)reg->addr);
4161 break;
4162 case KVM_REG_S390_EPOCHDIFF:
4163 r = get_user(vcpu->arch.sie_block->epoch,
4164 (u64 __user *)reg->addr);
4165 break;
4166 case KVM_REG_S390_CPU_TIMER:
4167 r = get_user(val, (u64 __user *)reg->addr);
4168 if (!r)
4169 kvm_s390_set_cpu_timer(vcpu, val);
4170 break;
4171 case KVM_REG_S390_CLOCK_COMP:
4172 r = get_user(vcpu->arch.sie_block->ckc,
4173 (u64 __user *)reg->addr);
4174 break;
4175 case KVM_REG_S390_PFTOKEN:
4176 r = get_user(vcpu->arch.pfault_token,
4177 (u64 __user *)reg->addr);
4178 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4179 kvm_clear_async_pf_completion_queue(vcpu);
4180 break;
4181 case KVM_REG_S390_PFCOMPARE:
4182 r = get_user(vcpu->arch.pfault_compare,
4183 (u64 __user *)reg->addr);
4184 break;
4185 case KVM_REG_S390_PFSELECT:
4186 r = get_user(vcpu->arch.pfault_select,
4187 (u64 __user *)reg->addr);
4188 break;
4189 case KVM_REG_S390_PP:
4190 r = get_user(vcpu->arch.sie_block->pp,
4191 (u64 __user *)reg->addr);
4192 break;
4193 case KVM_REG_S390_GBEA:
4194 r = get_user(vcpu->arch.sie_block->gbea,
4195 (u64 __user *)reg->addr);
4196 break;
4197 default:
4198 break;
4199 }
4200
4201 return r;
4202 }
4203
kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu * vcpu)4204 static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu)
4205 {
4206 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI;
4207 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
4208 memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb));
4209
4210 kvm_clear_async_pf_completion_queue(vcpu);
4211 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
4212 kvm_s390_vcpu_stop(vcpu);
4213 kvm_s390_clear_local_irqs(vcpu);
4214 }
4215
kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu * vcpu)4216 static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
4217 {
4218 /* Initial reset is a superset of the normal reset */
4219 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
4220
4221 /*
4222 * This equals initial cpu reset in pop, but we don't switch to ESA.
4223 * We do not only reset the internal data, but also ...
4224 */
4225 vcpu->arch.sie_block->gpsw.mask = 0;
4226 vcpu->arch.sie_block->gpsw.addr = 0;
4227 kvm_s390_set_prefix(vcpu, 0);
4228 kvm_s390_set_cpu_timer(vcpu, 0);
4229 vcpu->arch.sie_block->ckc = 0;
4230 memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr));
4231 vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK;
4232 vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK;
4233
4234 /* ... the data in sync regs */
4235 memset(vcpu->run->s.regs.crs, 0, sizeof(vcpu->run->s.regs.crs));
4236 vcpu->run->s.regs.ckc = 0;
4237 vcpu->run->s.regs.crs[0] = CR0_INITIAL_MASK;
4238 vcpu->run->s.regs.crs[14] = CR14_INITIAL_MASK;
4239 vcpu->run->psw_addr = 0;
4240 vcpu->run->psw_mask = 0;
4241 vcpu->run->s.regs.todpr = 0;
4242 vcpu->run->s.regs.cputm = 0;
4243 vcpu->run->s.regs.ckc = 0;
4244 vcpu->run->s.regs.pp = 0;
4245 vcpu->run->s.regs.gbea = 1;
4246 vcpu->run->s.regs.fpc = 0;
4247 /*
4248 * Do not reset these registers in the protected case, as some of
4249 * them are overlaid and they are not accessible in this case
4250 * anyway.
4251 */
4252 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4253 vcpu->arch.sie_block->gbea = 1;
4254 vcpu->arch.sie_block->pp = 0;
4255 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4256 vcpu->arch.sie_block->todpr = 0;
4257 }
4258 }
4259
kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu * vcpu)4260 static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu)
4261 {
4262 struct kvm_sync_regs *regs = &vcpu->run->s.regs;
4263
4264 /* Clear reset is a superset of the initial reset */
4265 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
4266
4267 memset(®s->gprs, 0, sizeof(regs->gprs));
4268 memset(®s->vrs, 0, sizeof(regs->vrs));
4269 memset(®s->acrs, 0, sizeof(regs->acrs));
4270 memset(®s->gscb, 0, sizeof(regs->gscb));
4271
4272 regs->etoken = 0;
4273 regs->etoken_extension = 0;
4274 }
4275
kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu * vcpu,struct kvm_regs * regs)4276 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4277 {
4278 vcpu_load(vcpu);
4279 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
4280 vcpu_put(vcpu);
4281 return 0;
4282 }
4283
kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu * vcpu,struct kvm_regs * regs)4284 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4285 {
4286 vcpu_load(vcpu);
4287 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
4288 vcpu_put(vcpu);
4289 return 0;
4290 }
4291
kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu * vcpu,struct kvm_sregs * sregs)4292 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4293 struct kvm_sregs *sregs)
4294 {
4295 vcpu_load(vcpu);
4296
4297 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
4298 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
4299
4300 vcpu_put(vcpu);
4301 return 0;
4302 }
4303
kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu * vcpu,struct kvm_sregs * sregs)4304 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4305 struct kvm_sregs *sregs)
4306 {
4307 vcpu_load(vcpu);
4308
4309 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
4310 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
4311
4312 vcpu_put(vcpu);
4313 return 0;
4314 }
4315
kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu * vcpu,struct kvm_fpu * fpu)4316 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4317 {
4318 int ret = 0;
4319
4320 vcpu_load(vcpu);
4321
4322 vcpu->run->s.regs.fpc = fpu->fpc;
4323 if (cpu_has_vx())
4324 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
4325 (freg_t *) fpu->fprs);
4326 else
4327 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
4328
4329 vcpu_put(vcpu);
4330 return ret;
4331 }
4332
kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu * vcpu,struct kvm_fpu * fpu)4333 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4334 {
4335 vcpu_load(vcpu);
4336
4337 if (cpu_has_vx())
4338 convert_vx_to_fp((freg_t *) fpu->fprs,
4339 (__vector128 *) vcpu->run->s.regs.vrs);
4340 else
4341 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
4342 fpu->fpc = vcpu->run->s.regs.fpc;
4343
4344 vcpu_put(vcpu);
4345 return 0;
4346 }
4347
kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu * vcpu,psw_t psw)4348 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
4349 {
4350 int rc = 0;
4351
4352 if (!is_vcpu_stopped(vcpu))
4353 rc = -EBUSY;
4354 else {
4355 vcpu->run->psw_mask = psw.mask;
4356 vcpu->run->psw_addr = psw.addr;
4357 }
4358 return rc;
4359 }
4360
kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu * vcpu,struct kvm_translation * tr)4361 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4362 struct kvm_translation *tr)
4363 {
4364 return -EINVAL; /* not implemented yet */
4365 }
4366
4367 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
4368 KVM_GUESTDBG_USE_HW_BP | \
4369 KVM_GUESTDBG_ENABLE)
4370
kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu * vcpu,struct kvm_guest_debug * dbg)4371 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4372 struct kvm_guest_debug *dbg)
4373 {
4374 int rc = 0;
4375
4376 vcpu_load(vcpu);
4377
4378 vcpu->guest_debug = 0;
4379 kvm_s390_clear_bp_data(vcpu);
4380
4381 if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
4382 rc = -EINVAL;
4383 goto out;
4384 }
4385 if (!sclp.has_gpere) {
4386 rc = -EINVAL;
4387 goto out;
4388 }
4389
4390 if (dbg->control & KVM_GUESTDBG_ENABLE) {
4391 vcpu->guest_debug = dbg->control;
4392 /* enforce guest PER */
4393 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
4394
4395 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
4396 rc = kvm_s390_import_bp_data(vcpu, dbg);
4397 } else {
4398 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4399 vcpu->arch.guestdbg.last_bp = 0;
4400 }
4401
4402 if (rc) {
4403 vcpu->guest_debug = 0;
4404 kvm_s390_clear_bp_data(vcpu);
4405 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4406 }
4407
4408 out:
4409 vcpu_put(vcpu);
4410 return rc;
4411 }
4412
kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu * vcpu,struct kvm_mp_state * mp_state)4413 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4414 struct kvm_mp_state *mp_state)
4415 {
4416 int ret;
4417
4418 vcpu_load(vcpu);
4419
4420 /* CHECK_STOP and LOAD are not supported yet */
4421 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
4422 KVM_MP_STATE_OPERATING;
4423
4424 vcpu_put(vcpu);
4425 return ret;
4426 }
4427
kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu * vcpu,struct kvm_mp_state * mp_state)4428 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4429 struct kvm_mp_state *mp_state)
4430 {
4431 int rc = 0;
4432
4433 vcpu_load(vcpu);
4434
4435 /* user space knows about this interface - let it control the state */
4436 kvm_s390_set_user_cpu_state_ctrl(vcpu->kvm);
4437
4438 switch (mp_state->mp_state) {
4439 case KVM_MP_STATE_STOPPED:
4440 rc = kvm_s390_vcpu_stop(vcpu);
4441 break;
4442 case KVM_MP_STATE_OPERATING:
4443 rc = kvm_s390_vcpu_start(vcpu);
4444 break;
4445 case KVM_MP_STATE_LOAD:
4446 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4447 rc = -ENXIO;
4448 break;
4449 }
4450 rc = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR_LOAD);
4451 break;
4452 case KVM_MP_STATE_CHECK_STOP:
4453 fallthrough; /* CHECK_STOP and LOAD are not supported yet */
4454 default:
4455 rc = -ENXIO;
4456 }
4457
4458 vcpu_put(vcpu);
4459 return rc;
4460 }
4461
ibs_enabled(struct kvm_vcpu * vcpu)4462 static bool ibs_enabled(struct kvm_vcpu *vcpu)
4463 {
4464 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
4465 }
4466
kvm_s390_handle_requests(struct kvm_vcpu * vcpu)4467 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
4468 {
4469 retry:
4470 kvm_s390_vcpu_request_handled(vcpu);
4471 if (!kvm_request_pending(vcpu))
4472 return 0;
4473 /*
4474 * If the guest prefix changed, re-arm the ipte notifier for the
4475 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
4476 * This ensures that the ipte instruction for this request has
4477 * already finished. We might race against a second unmapper that
4478 * wants to set the blocking bit. Lets just retry the request loop.
4479 */
4480 if (kvm_check_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu)) {
4481 int rc;
4482 rc = gmap_mprotect_notify(vcpu->arch.gmap,
4483 kvm_s390_get_prefix(vcpu),
4484 PAGE_SIZE * 2, PROT_WRITE);
4485 if (rc) {
4486 kvm_make_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4487 return rc;
4488 }
4489 goto retry;
4490 }
4491
4492 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
4493 vcpu->arch.sie_block->ihcpu = 0xffff;
4494 goto retry;
4495 }
4496
4497 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
4498 if (!ibs_enabled(vcpu)) {
4499 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
4500 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
4501 }
4502 goto retry;
4503 }
4504
4505 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
4506 if (ibs_enabled(vcpu)) {
4507 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
4508 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
4509 }
4510 goto retry;
4511 }
4512
4513 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
4514 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
4515 goto retry;
4516 }
4517
4518 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
4519 /*
4520 * Disable CMM virtualization; we will emulate the ESSA
4521 * instruction manually, in order to provide additional
4522 * functionalities needed for live migration.
4523 */
4524 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
4525 goto retry;
4526 }
4527
4528 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
4529 /*
4530 * Re-enable CMM virtualization if CMMA is available and
4531 * CMM has been used.
4532 */
4533 if ((vcpu->kvm->arch.use_cmma) &&
4534 (vcpu->kvm->mm->context.uses_cmm))
4535 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
4536 goto retry;
4537 }
4538
4539 /* we left the vsie handler, nothing to do, just clear the request */
4540 kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu);
4541
4542 return 0;
4543 }
4544
__kvm_s390_set_tod_clock(struct kvm * kvm,const struct kvm_s390_vm_tod_clock * gtod)4545 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4546 {
4547 struct kvm_vcpu *vcpu;
4548 union tod_clock clk;
4549 unsigned long i;
4550
4551 preempt_disable();
4552
4553 store_tod_clock_ext(&clk);
4554
4555 kvm->arch.epoch = gtod->tod - clk.tod;
4556 kvm->arch.epdx = 0;
4557 if (test_kvm_facility(kvm, 139)) {
4558 kvm->arch.epdx = gtod->epoch_idx - clk.ei;
4559 if (kvm->arch.epoch > gtod->tod)
4560 kvm->arch.epdx -= 1;
4561 }
4562
4563 kvm_s390_vcpu_block_all(kvm);
4564 kvm_for_each_vcpu(i, vcpu, kvm) {
4565 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
4566 vcpu->arch.sie_block->epdx = kvm->arch.epdx;
4567 }
4568
4569 kvm_s390_vcpu_unblock_all(kvm);
4570 preempt_enable();
4571 }
4572
kvm_s390_try_set_tod_clock(struct kvm * kvm,const struct kvm_s390_vm_tod_clock * gtod)4573 int kvm_s390_try_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4574 {
4575 if (!mutex_trylock(&kvm->lock))
4576 return 0;
4577 __kvm_s390_set_tod_clock(kvm, gtod);
4578 mutex_unlock(&kvm->lock);
4579 return 1;
4580 }
4581
4582 /**
4583 * kvm_arch_fault_in_page - fault-in guest page if necessary
4584 * @vcpu: The corresponding virtual cpu
4585 * @gpa: Guest physical address
4586 * @writable: Whether the page should be writable or not
4587 *
4588 * Make sure that a guest page has been faulted-in on the host.
4589 *
4590 * Return: Zero on success, negative error code otherwise.
4591 */
kvm_arch_fault_in_page(struct kvm_vcpu * vcpu,gpa_t gpa,int writable)4592 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
4593 {
4594 return gmap_fault(vcpu->arch.gmap, gpa,
4595 writable ? FAULT_FLAG_WRITE : 0);
4596 }
4597
__kvm_inject_pfault_token(struct kvm_vcpu * vcpu,bool start_token,unsigned long token)4598 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
4599 unsigned long token)
4600 {
4601 struct kvm_s390_interrupt inti;
4602 struct kvm_s390_irq irq;
4603
4604 if (start_token) {
4605 irq.u.ext.ext_params2 = token;
4606 irq.type = KVM_S390_INT_PFAULT_INIT;
4607 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
4608 } else {
4609 inti.type = KVM_S390_INT_PFAULT_DONE;
4610 inti.parm64 = token;
4611 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
4612 }
4613 }
4614
kvm_arch_async_page_not_present(struct kvm_vcpu * vcpu,struct kvm_async_pf * work)4615 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
4616 struct kvm_async_pf *work)
4617 {
4618 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
4619 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
4620
4621 return true;
4622 }
4623
kvm_arch_async_page_present(struct kvm_vcpu * vcpu,struct kvm_async_pf * work)4624 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
4625 struct kvm_async_pf *work)
4626 {
4627 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
4628 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
4629 }
4630
kvm_arch_async_page_ready(struct kvm_vcpu * vcpu,struct kvm_async_pf * work)4631 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
4632 struct kvm_async_pf *work)
4633 {
4634 /* s390 will always inject the page directly */
4635 }
4636
kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu * vcpu)4637 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
4638 {
4639 /*
4640 * s390 will always inject the page directly,
4641 * but we still want check_async_completion to cleanup
4642 */
4643 return true;
4644 }
4645
kvm_arch_setup_async_pf(struct kvm_vcpu * vcpu)4646 static bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
4647 {
4648 hva_t hva;
4649 struct kvm_arch_async_pf arch;
4650
4651 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4652 return false;
4653 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
4654 vcpu->arch.pfault_compare)
4655 return false;
4656 if (psw_extint_disabled(vcpu))
4657 return false;
4658 if (kvm_s390_vcpu_has_irq(vcpu, 0))
4659 return false;
4660 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
4661 return false;
4662 if (!vcpu->arch.gmap->pfault_enabled)
4663 return false;
4664
4665 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
4666 hva += current->thread.gmap_addr & ~PAGE_MASK;
4667 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
4668 return false;
4669
4670 return kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
4671 }
4672
vcpu_pre_run(struct kvm_vcpu * vcpu)4673 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
4674 {
4675 int rc, cpuflags;
4676
4677 /*
4678 * On s390 notifications for arriving pages will be delivered directly
4679 * to the guest but the house keeping for completed pfaults is
4680 * handled outside the worker.
4681 */
4682 kvm_check_async_pf_completion(vcpu);
4683
4684 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
4685 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
4686
4687 if (need_resched())
4688 schedule();
4689
4690 if (!kvm_is_ucontrol(vcpu->kvm)) {
4691 rc = kvm_s390_deliver_pending_interrupts(vcpu);
4692 if (rc || guestdbg_exit_pending(vcpu))
4693 return rc;
4694 }
4695
4696 rc = kvm_s390_handle_requests(vcpu);
4697 if (rc)
4698 return rc;
4699
4700 if (guestdbg_enabled(vcpu)) {
4701 kvm_s390_backup_guest_per_regs(vcpu);
4702 kvm_s390_patch_guest_per_regs(vcpu);
4703 }
4704
4705 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4706
4707 vcpu->arch.sie_block->icptcode = 0;
4708 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
4709 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
4710 trace_kvm_s390_sie_enter(vcpu, cpuflags);
4711
4712 return 0;
4713 }
4714
vcpu_post_run_fault_in_sie(struct kvm_vcpu * vcpu)4715 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
4716 {
4717 struct kvm_s390_pgm_info pgm_info = {
4718 .code = PGM_ADDRESSING,
4719 };
4720 u8 opcode, ilen;
4721 int rc;
4722
4723 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
4724 trace_kvm_s390_sie_fault(vcpu);
4725
4726 /*
4727 * We want to inject an addressing exception, which is defined as a
4728 * suppressing or terminating exception. However, since we came here
4729 * by a DAT access exception, the PSW still points to the faulting
4730 * instruction since DAT exceptions are nullifying. So we've got
4731 * to look up the current opcode to get the length of the instruction
4732 * to be able to forward the PSW.
4733 */
4734 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
4735 ilen = insn_length(opcode);
4736 if (rc < 0) {
4737 return rc;
4738 } else if (rc) {
4739 /* Instruction-Fetching Exceptions - we can't detect the ilen.
4740 * Forward by arbitrary ilc, injection will take care of
4741 * nullification if necessary.
4742 */
4743 pgm_info = vcpu->arch.pgm;
4744 ilen = 4;
4745 }
4746 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
4747 kvm_s390_forward_psw(vcpu, ilen);
4748 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
4749 }
4750
vcpu_post_run(struct kvm_vcpu * vcpu,int exit_reason)4751 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
4752 {
4753 struct mcck_volatile_info *mcck_info;
4754 struct sie_page *sie_page;
4755
4756 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
4757 vcpu->arch.sie_block->icptcode);
4758 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
4759
4760 if (guestdbg_enabled(vcpu))
4761 kvm_s390_restore_guest_per_regs(vcpu);
4762
4763 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
4764 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
4765
4766 if (exit_reason == -EINTR) {
4767 VCPU_EVENT(vcpu, 3, "%s", "machine check");
4768 sie_page = container_of(vcpu->arch.sie_block,
4769 struct sie_page, sie_block);
4770 mcck_info = &sie_page->mcck_info;
4771 kvm_s390_reinject_machine_check(vcpu, mcck_info);
4772 return 0;
4773 }
4774
4775 if (vcpu->arch.sie_block->icptcode > 0) {
4776 int rc = kvm_handle_sie_intercept(vcpu);
4777
4778 if (rc != -EOPNOTSUPP)
4779 return rc;
4780 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
4781 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
4782 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
4783 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
4784 return -EREMOTE;
4785 } else if (exit_reason != -EFAULT) {
4786 vcpu->stat.exit_null++;
4787 return 0;
4788 } else if (kvm_is_ucontrol(vcpu->kvm)) {
4789 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
4790 vcpu->run->s390_ucontrol.trans_exc_code =
4791 current->thread.gmap_addr;
4792 vcpu->run->s390_ucontrol.pgm_code = 0x10;
4793 return -EREMOTE;
4794 } else if (current->thread.gmap_pfault) {
4795 trace_kvm_s390_major_guest_pfault(vcpu);
4796 current->thread.gmap_pfault = 0;
4797 if (kvm_arch_setup_async_pf(vcpu))
4798 return 0;
4799 vcpu->stat.pfault_sync++;
4800 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
4801 }
4802 return vcpu_post_run_fault_in_sie(vcpu);
4803 }
4804
4805 #define PSW_INT_MASK (PSW_MASK_EXT | PSW_MASK_IO | PSW_MASK_MCHECK)
__vcpu_run(struct kvm_vcpu * vcpu)4806 static int __vcpu_run(struct kvm_vcpu *vcpu)
4807 {
4808 int rc, exit_reason;
4809 struct sie_page *sie_page = (struct sie_page *)vcpu->arch.sie_block;
4810
4811 /*
4812 * We try to hold kvm->srcu during most of vcpu_run (except when run-
4813 * ning the guest), so that memslots (and other stuff) are protected
4814 */
4815 kvm_vcpu_srcu_read_lock(vcpu);
4816
4817 do {
4818 rc = vcpu_pre_run(vcpu);
4819 if (rc || guestdbg_exit_pending(vcpu))
4820 break;
4821
4822 kvm_vcpu_srcu_read_unlock(vcpu);
4823 /*
4824 * As PF_VCPU will be used in fault handler, between
4825 * guest_enter and guest_exit should be no uaccess.
4826 */
4827 local_irq_disable();
4828 guest_enter_irqoff();
4829 __disable_cpu_timer_accounting(vcpu);
4830 local_irq_enable();
4831 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4832 memcpy(sie_page->pv_grregs,
4833 vcpu->run->s.regs.gprs,
4834 sizeof(sie_page->pv_grregs));
4835 }
4836 exit_reason = sie64a(vcpu->arch.sie_block,
4837 vcpu->run->s.regs.gprs,
4838 gmap_get_enabled()->asce);
4839 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4840 memcpy(vcpu->run->s.regs.gprs,
4841 sie_page->pv_grregs,
4842 sizeof(sie_page->pv_grregs));
4843 /*
4844 * We're not allowed to inject interrupts on intercepts
4845 * that leave the guest state in an "in-between" state
4846 * where the next SIE entry will do a continuation.
4847 * Fence interrupts in our "internal" PSW.
4848 */
4849 if (vcpu->arch.sie_block->icptcode == ICPT_PV_INSTR ||
4850 vcpu->arch.sie_block->icptcode == ICPT_PV_PREF) {
4851 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
4852 }
4853 }
4854 local_irq_disable();
4855 __enable_cpu_timer_accounting(vcpu);
4856 guest_exit_irqoff();
4857 local_irq_enable();
4858 kvm_vcpu_srcu_read_lock(vcpu);
4859
4860 rc = vcpu_post_run(vcpu, exit_reason);
4861 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
4862
4863 kvm_vcpu_srcu_read_unlock(vcpu);
4864 return rc;
4865 }
4866
sync_regs_fmt2(struct kvm_vcpu * vcpu)4867 static void sync_regs_fmt2(struct kvm_vcpu *vcpu)
4868 {
4869 struct kvm_run *kvm_run = vcpu->run;
4870 struct runtime_instr_cb *riccb;
4871 struct gs_cb *gscb;
4872
4873 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
4874 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
4875 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
4876 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
4877 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4878 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
4879 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
4880 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
4881 }
4882 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
4883 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
4884 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
4885 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
4886 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4887 kvm_clear_async_pf_completion_queue(vcpu);
4888 }
4889 if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) {
4890 vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318;
4891 vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc;
4892 VCPU_EVENT(vcpu, 3, "setting cpnc to %d", vcpu->arch.diag318_info.cpnc);
4893 }
4894 /*
4895 * If userspace sets the riccb (e.g. after migration) to a valid state,
4896 * we should enable RI here instead of doing the lazy enablement.
4897 */
4898 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
4899 test_kvm_facility(vcpu->kvm, 64) &&
4900 riccb->v &&
4901 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
4902 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
4903 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
4904 }
4905 /*
4906 * If userspace sets the gscb (e.g. after migration) to non-zero,
4907 * we should enable GS here instead of doing the lazy enablement.
4908 */
4909 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
4910 test_kvm_facility(vcpu->kvm, 133) &&
4911 gscb->gssm &&
4912 !vcpu->arch.gs_enabled) {
4913 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
4914 vcpu->arch.sie_block->ecb |= ECB_GS;
4915 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
4916 vcpu->arch.gs_enabled = 1;
4917 }
4918 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
4919 test_kvm_facility(vcpu->kvm, 82)) {
4920 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4921 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
4922 }
4923 if (MACHINE_HAS_GS) {
4924 preempt_disable();
4925 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
4926 if (current->thread.gs_cb) {
4927 vcpu->arch.host_gscb = current->thread.gs_cb;
4928 save_gs_cb(vcpu->arch.host_gscb);
4929 }
4930 if (vcpu->arch.gs_enabled) {
4931 current->thread.gs_cb = (struct gs_cb *)
4932 &vcpu->run->s.regs.gscb;
4933 restore_gs_cb(current->thread.gs_cb);
4934 }
4935 preempt_enable();
4936 }
4937 /* SIE will load etoken directly from SDNX and therefore kvm_run */
4938 }
4939
sync_regs(struct kvm_vcpu * vcpu)4940 static void sync_regs(struct kvm_vcpu *vcpu)
4941 {
4942 struct kvm_run *kvm_run = vcpu->run;
4943
4944 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
4945 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
4946 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
4947 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
4948 /* some control register changes require a tlb flush */
4949 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4950 }
4951 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4952 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
4953 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
4954 }
4955 save_access_regs(vcpu->arch.host_acrs);
4956 restore_access_regs(vcpu->run->s.regs.acrs);
4957 vcpu->arch.acrs_loaded = true;
4958 kvm_s390_fpu_load(vcpu->run);
4959 /* Sync fmt2 only data */
4960 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) {
4961 sync_regs_fmt2(vcpu);
4962 } else {
4963 /*
4964 * In several places we have to modify our internal view to
4965 * not do things that are disallowed by the ultravisor. For
4966 * example we must not inject interrupts after specific exits
4967 * (e.g. 112 prefix page not secure). We do this by turning
4968 * off the machine check, external and I/O interrupt bits
4969 * of our PSW copy. To avoid getting validity intercepts, we
4970 * do only accept the condition code from userspace.
4971 */
4972 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_CC;
4973 vcpu->arch.sie_block->gpsw.mask |= kvm_run->psw_mask &
4974 PSW_MASK_CC;
4975 }
4976
4977 kvm_run->kvm_dirty_regs = 0;
4978 }
4979
store_regs_fmt2(struct kvm_vcpu * vcpu)4980 static void store_regs_fmt2(struct kvm_vcpu *vcpu)
4981 {
4982 struct kvm_run *kvm_run = vcpu->run;
4983
4984 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
4985 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
4986 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
4987 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
4988 kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val;
4989 if (MACHINE_HAS_GS) {
4990 preempt_disable();
4991 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
4992 if (vcpu->arch.gs_enabled)
4993 save_gs_cb(current->thread.gs_cb);
4994 current->thread.gs_cb = vcpu->arch.host_gscb;
4995 restore_gs_cb(vcpu->arch.host_gscb);
4996 if (!vcpu->arch.host_gscb)
4997 local_ctl_clear_bit(2, CR2_GUARDED_STORAGE_BIT);
4998 vcpu->arch.host_gscb = NULL;
4999 preempt_enable();
5000 }
5001 /* SIE will save etoken directly into SDNX and therefore kvm_run */
5002 }
5003
store_regs(struct kvm_vcpu * vcpu)5004 static void store_regs(struct kvm_vcpu *vcpu)
5005 {
5006 struct kvm_run *kvm_run = vcpu->run;
5007
5008 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
5009 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
5010 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
5011 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
5012 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
5013 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
5014 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
5015 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
5016 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
5017 save_access_regs(vcpu->run->s.regs.acrs);
5018 restore_access_regs(vcpu->arch.host_acrs);
5019 vcpu->arch.acrs_loaded = false;
5020 kvm_s390_fpu_store(vcpu->run);
5021 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu)))
5022 store_regs_fmt2(vcpu);
5023 }
5024
kvm_arch_vcpu_ioctl_run(struct kvm_vcpu * vcpu)5025 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
5026 {
5027 struct kvm_run *kvm_run = vcpu->run;
5028 DECLARE_KERNEL_FPU_ONSTACK32(fpu);
5029 int rc;
5030
5031 /*
5032 * Running a VM while dumping always has the potential to
5033 * produce inconsistent dump data. But for PV vcpus a SIE
5034 * entry while dumping could also lead to a fatal validity
5035 * intercept which we absolutely want to avoid.
5036 */
5037 if (vcpu->kvm->arch.pv.dumping)
5038 return -EINVAL;
5039
5040 if (!vcpu->wants_to_run)
5041 return -EINTR;
5042
5043 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS ||
5044 kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS)
5045 return -EINVAL;
5046
5047 vcpu_load(vcpu);
5048
5049 if (guestdbg_exit_pending(vcpu)) {
5050 kvm_s390_prepare_debug_exit(vcpu);
5051 rc = 0;
5052 goto out;
5053 }
5054
5055 kvm_sigset_activate(vcpu);
5056
5057 /*
5058 * no need to check the return value of vcpu_start as it can only have
5059 * an error for protvirt, but protvirt means user cpu state
5060 */
5061 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
5062 kvm_s390_vcpu_start(vcpu);
5063 } else if (is_vcpu_stopped(vcpu)) {
5064 pr_err_ratelimited("can't run stopped vcpu %d\n",
5065 vcpu->vcpu_id);
5066 rc = -EINVAL;
5067 goto out;
5068 }
5069
5070 kernel_fpu_begin(&fpu, KERNEL_FPC | KERNEL_VXR);
5071 sync_regs(vcpu);
5072 enable_cpu_timer_accounting(vcpu);
5073
5074 might_fault();
5075 rc = __vcpu_run(vcpu);
5076
5077 if (signal_pending(current) && !rc) {
5078 kvm_run->exit_reason = KVM_EXIT_INTR;
5079 rc = -EINTR;
5080 }
5081
5082 if (guestdbg_exit_pending(vcpu) && !rc) {
5083 kvm_s390_prepare_debug_exit(vcpu);
5084 rc = 0;
5085 }
5086
5087 if (rc == -EREMOTE) {
5088 /* userspace support is needed, kvm_run has been prepared */
5089 rc = 0;
5090 }
5091
5092 disable_cpu_timer_accounting(vcpu);
5093 store_regs(vcpu);
5094 kernel_fpu_end(&fpu, KERNEL_FPC | KERNEL_VXR);
5095
5096 kvm_sigset_deactivate(vcpu);
5097
5098 vcpu->stat.exit_userspace++;
5099 out:
5100 vcpu_put(vcpu);
5101 return rc;
5102 }
5103
5104 /*
5105 * store status at address
5106 * we use have two special cases:
5107 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
5108 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
5109 */
kvm_s390_store_status_unloaded(struct kvm_vcpu * vcpu,unsigned long gpa)5110 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
5111 {
5112 unsigned char archmode = 1;
5113 freg_t fprs[NUM_FPRS];
5114 unsigned int px;
5115 u64 clkcomp, cputm;
5116 int rc;
5117
5118 px = kvm_s390_get_prefix(vcpu);
5119 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
5120 if (write_guest_abs(vcpu, 163, &archmode, 1))
5121 return -EFAULT;
5122 gpa = 0;
5123 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
5124 if (write_guest_real(vcpu, 163, &archmode, 1))
5125 return -EFAULT;
5126 gpa = px;
5127 } else
5128 gpa -= __LC_FPREGS_SAVE_AREA;
5129
5130 /* manually convert vector registers if necessary */
5131 if (cpu_has_vx()) {
5132 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
5133 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5134 fprs, 128);
5135 } else {
5136 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5137 vcpu->run->s.regs.fprs, 128);
5138 }
5139 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
5140 vcpu->run->s.regs.gprs, 128);
5141 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
5142 &vcpu->arch.sie_block->gpsw, 16);
5143 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
5144 &px, 4);
5145 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
5146 &vcpu->run->s.regs.fpc, 4);
5147 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
5148 &vcpu->arch.sie_block->todpr, 4);
5149 cputm = kvm_s390_get_cpu_timer(vcpu);
5150 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
5151 &cputm, 8);
5152 clkcomp = vcpu->arch.sie_block->ckc >> 8;
5153 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
5154 &clkcomp, 8);
5155 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
5156 &vcpu->run->s.regs.acrs, 64);
5157 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
5158 &vcpu->arch.sie_block->gcr, 128);
5159 return rc ? -EFAULT : 0;
5160 }
5161
kvm_s390_vcpu_store_status(struct kvm_vcpu * vcpu,unsigned long addr)5162 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
5163 {
5164 /*
5165 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
5166 * switch in the run ioctl. Let's update our copies before we save
5167 * it into the save area
5168 */
5169 kvm_s390_fpu_store(vcpu->run);
5170 save_access_regs(vcpu->run->s.regs.acrs);
5171
5172 return kvm_s390_store_status_unloaded(vcpu, addr);
5173 }
5174
__disable_ibs_on_vcpu(struct kvm_vcpu * vcpu)5175 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5176 {
5177 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
5178 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
5179 }
5180
__disable_ibs_on_all_vcpus(struct kvm * kvm)5181 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
5182 {
5183 unsigned long i;
5184 struct kvm_vcpu *vcpu;
5185
5186 kvm_for_each_vcpu(i, vcpu, kvm) {
5187 __disable_ibs_on_vcpu(vcpu);
5188 }
5189 }
5190
__enable_ibs_on_vcpu(struct kvm_vcpu * vcpu)5191 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5192 {
5193 if (!sclp.has_ibs)
5194 return;
5195 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
5196 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
5197 }
5198
kvm_s390_vcpu_start(struct kvm_vcpu * vcpu)5199 int kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
5200 {
5201 int i, online_vcpus, r = 0, started_vcpus = 0;
5202
5203 if (!is_vcpu_stopped(vcpu))
5204 return 0;
5205
5206 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
5207 /* Only one cpu at a time may enter/leave the STOPPED state. */
5208 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5209 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5210
5211 /* Let's tell the UV that we want to change into the operating state */
5212 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5213 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR);
5214 if (r) {
5215 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5216 return r;
5217 }
5218 }
5219
5220 for (i = 0; i < online_vcpus; i++) {
5221 if (!is_vcpu_stopped(kvm_get_vcpu(vcpu->kvm, i)))
5222 started_vcpus++;
5223 }
5224
5225 if (started_vcpus == 0) {
5226 /* we're the only active VCPU -> speed it up */
5227 __enable_ibs_on_vcpu(vcpu);
5228 } else if (started_vcpus == 1) {
5229 /*
5230 * As we are starting a second VCPU, we have to disable
5231 * the IBS facility on all VCPUs to remove potentially
5232 * outstanding ENABLE requests.
5233 */
5234 __disable_ibs_on_all_vcpus(vcpu->kvm);
5235 }
5236
5237 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
5238 /*
5239 * The real PSW might have changed due to a RESTART interpreted by the
5240 * ultravisor. We block all interrupts and let the next sie exit
5241 * refresh our view.
5242 */
5243 if (kvm_s390_pv_cpu_is_protected(vcpu))
5244 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
5245 /*
5246 * Another VCPU might have used IBS while we were offline.
5247 * Let's play safe and flush the VCPU at startup.
5248 */
5249 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5250 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5251 return 0;
5252 }
5253
kvm_s390_vcpu_stop(struct kvm_vcpu * vcpu)5254 int kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
5255 {
5256 int i, online_vcpus, r = 0, started_vcpus = 0;
5257 struct kvm_vcpu *started_vcpu = NULL;
5258
5259 if (is_vcpu_stopped(vcpu))
5260 return 0;
5261
5262 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
5263 /* Only one cpu at a time may enter/leave the STOPPED state. */
5264 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5265 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5266
5267 /* Let's tell the UV that we want to change into the stopped state */
5268 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5269 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_STP);
5270 if (r) {
5271 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5272 return r;
5273 }
5274 }
5275
5276 /*
5277 * Set the VCPU to STOPPED and THEN clear the interrupt flag,
5278 * now that the SIGP STOP and SIGP STOP AND STORE STATUS orders
5279 * have been fully processed. This will ensure that the VCPU
5280 * is kept BUSY if another VCPU is inquiring with SIGP SENSE.
5281 */
5282 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
5283 kvm_s390_clear_stop_irq(vcpu);
5284
5285 __disable_ibs_on_vcpu(vcpu);
5286
5287 for (i = 0; i < online_vcpus; i++) {
5288 struct kvm_vcpu *tmp = kvm_get_vcpu(vcpu->kvm, i);
5289
5290 if (!is_vcpu_stopped(tmp)) {
5291 started_vcpus++;
5292 started_vcpu = tmp;
5293 }
5294 }
5295
5296 if (started_vcpus == 1) {
5297 /*
5298 * As we only have one VCPU left, we want to enable the
5299 * IBS facility for that VCPU to speed it up.
5300 */
5301 __enable_ibs_on_vcpu(started_vcpu);
5302 }
5303
5304 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5305 return 0;
5306 }
5307
kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu * vcpu,struct kvm_enable_cap * cap)5308 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5309 struct kvm_enable_cap *cap)
5310 {
5311 int r;
5312
5313 if (cap->flags)
5314 return -EINVAL;
5315
5316 switch (cap->cap) {
5317 case KVM_CAP_S390_CSS_SUPPORT:
5318 if (!vcpu->kvm->arch.css_support) {
5319 vcpu->kvm->arch.css_support = 1;
5320 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
5321 trace_kvm_s390_enable_css(vcpu->kvm);
5322 }
5323 r = 0;
5324 break;
5325 default:
5326 r = -EINVAL;
5327 break;
5328 }
5329 return r;
5330 }
5331
kvm_s390_vcpu_sida_op(struct kvm_vcpu * vcpu,struct kvm_s390_mem_op * mop)5332 static long kvm_s390_vcpu_sida_op(struct kvm_vcpu *vcpu,
5333 struct kvm_s390_mem_op *mop)
5334 {
5335 void __user *uaddr = (void __user *)mop->buf;
5336 void *sida_addr;
5337 int r = 0;
5338
5339 if (mop->flags || !mop->size)
5340 return -EINVAL;
5341 if (mop->size + mop->sida_offset < mop->size)
5342 return -EINVAL;
5343 if (mop->size + mop->sida_offset > sida_size(vcpu->arch.sie_block))
5344 return -E2BIG;
5345 if (!kvm_s390_pv_cpu_is_protected(vcpu))
5346 return -EINVAL;
5347
5348 sida_addr = (char *)sida_addr(vcpu->arch.sie_block) + mop->sida_offset;
5349
5350 switch (mop->op) {
5351 case KVM_S390_MEMOP_SIDA_READ:
5352 if (copy_to_user(uaddr, sida_addr, mop->size))
5353 r = -EFAULT;
5354
5355 break;
5356 case KVM_S390_MEMOP_SIDA_WRITE:
5357 if (copy_from_user(sida_addr, uaddr, mop->size))
5358 r = -EFAULT;
5359 break;
5360 }
5361 return r;
5362 }
5363
kvm_s390_vcpu_mem_op(struct kvm_vcpu * vcpu,struct kvm_s390_mem_op * mop)5364 static long kvm_s390_vcpu_mem_op(struct kvm_vcpu *vcpu,
5365 struct kvm_s390_mem_op *mop)
5366 {
5367 void __user *uaddr = (void __user *)mop->buf;
5368 enum gacc_mode acc_mode;
5369 void *tmpbuf = NULL;
5370 int r;
5371
5372 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_INJECT_EXCEPTION |
5373 KVM_S390_MEMOP_F_CHECK_ONLY |
5374 KVM_S390_MEMOP_F_SKEY_PROTECTION);
5375 if (r)
5376 return r;
5377 if (mop->ar >= NUM_ACRS)
5378 return -EINVAL;
5379 if (kvm_s390_pv_cpu_is_protected(vcpu))
5380 return -EINVAL;
5381 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
5382 tmpbuf = vmalloc(mop->size);
5383 if (!tmpbuf)
5384 return -ENOMEM;
5385 }
5386
5387 acc_mode = mop->op == KVM_S390_MEMOP_LOGICAL_READ ? GACC_FETCH : GACC_STORE;
5388 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
5389 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size,
5390 acc_mode, mop->key);
5391 goto out_inject;
5392 }
5393 if (acc_mode == GACC_FETCH) {
5394 r = read_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5395 mop->size, mop->key);
5396 if (r)
5397 goto out_inject;
5398 if (copy_to_user(uaddr, tmpbuf, mop->size)) {
5399 r = -EFAULT;
5400 goto out_free;
5401 }
5402 } else {
5403 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
5404 r = -EFAULT;
5405 goto out_free;
5406 }
5407 r = write_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5408 mop->size, mop->key);
5409 }
5410
5411 out_inject:
5412 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
5413 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
5414
5415 out_free:
5416 vfree(tmpbuf);
5417 return r;
5418 }
5419
kvm_s390_vcpu_memsida_op(struct kvm_vcpu * vcpu,struct kvm_s390_mem_op * mop)5420 static long kvm_s390_vcpu_memsida_op(struct kvm_vcpu *vcpu,
5421 struct kvm_s390_mem_op *mop)
5422 {
5423 int r, srcu_idx;
5424
5425 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5426
5427 switch (mop->op) {
5428 case KVM_S390_MEMOP_LOGICAL_READ:
5429 case KVM_S390_MEMOP_LOGICAL_WRITE:
5430 r = kvm_s390_vcpu_mem_op(vcpu, mop);
5431 break;
5432 case KVM_S390_MEMOP_SIDA_READ:
5433 case KVM_S390_MEMOP_SIDA_WRITE:
5434 /* we are locked against sida going away by the vcpu->mutex */
5435 r = kvm_s390_vcpu_sida_op(vcpu, mop);
5436 break;
5437 default:
5438 r = -EINVAL;
5439 }
5440
5441 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
5442 return r;
5443 }
5444
kvm_arch_vcpu_async_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)5445 long kvm_arch_vcpu_async_ioctl(struct file *filp,
5446 unsigned int ioctl, unsigned long arg)
5447 {
5448 struct kvm_vcpu *vcpu = filp->private_data;
5449 void __user *argp = (void __user *)arg;
5450 int rc;
5451
5452 switch (ioctl) {
5453 case KVM_S390_IRQ: {
5454 struct kvm_s390_irq s390irq;
5455
5456 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
5457 return -EFAULT;
5458 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5459 break;
5460 }
5461 case KVM_S390_INTERRUPT: {
5462 struct kvm_s390_interrupt s390int;
5463 struct kvm_s390_irq s390irq = {};
5464
5465 if (copy_from_user(&s390int, argp, sizeof(s390int)))
5466 return -EFAULT;
5467 if (s390int_to_s390irq(&s390int, &s390irq))
5468 return -EINVAL;
5469 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5470 break;
5471 }
5472 default:
5473 rc = -ENOIOCTLCMD;
5474 break;
5475 }
5476
5477 /*
5478 * To simplify single stepping of userspace-emulated instructions,
5479 * KVM_EXIT_S390_SIEIC exit sets KVM_GUESTDBG_EXIT_PENDING (see
5480 * should_handle_per_ifetch()). However, if userspace emulation injects
5481 * an interrupt, it needs to be cleared, so that KVM_EXIT_DEBUG happens
5482 * after (and not before) the interrupt delivery.
5483 */
5484 if (!rc)
5485 vcpu->guest_debug &= ~KVM_GUESTDBG_EXIT_PENDING;
5486
5487 return rc;
5488 }
5489
kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu * vcpu,struct kvm_pv_cmd * cmd)5490 static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu,
5491 struct kvm_pv_cmd *cmd)
5492 {
5493 struct kvm_s390_pv_dmp dmp;
5494 void *data;
5495 int ret;
5496
5497 /* Dump initialization is a prerequisite */
5498 if (!vcpu->kvm->arch.pv.dumping)
5499 return -EINVAL;
5500
5501 if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp)))
5502 return -EFAULT;
5503
5504 /* We only handle this subcmd right now */
5505 if (dmp.subcmd != KVM_PV_DUMP_CPU)
5506 return -EINVAL;
5507
5508 /* CPU dump length is the same as create cpu storage donation. */
5509 if (dmp.buff_len != uv_info.guest_cpu_stor_len)
5510 return -EINVAL;
5511
5512 data = kvzalloc(uv_info.guest_cpu_stor_len, GFP_KERNEL);
5513 if (!data)
5514 return -ENOMEM;
5515
5516 ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc);
5517
5518 VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x",
5519 vcpu->vcpu_id, cmd->rc, cmd->rrc);
5520
5521 if (ret)
5522 ret = -EINVAL;
5523
5524 /* On success copy over the dump data */
5525 if (!ret && copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len))
5526 ret = -EFAULT;
5527
5528 kvfree(data);
5529 return ret;
5530 }
5531
kvm_arch_vcpu_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)5532 long kvm_arch_vcpu_ioctl(struct file *filp,
5533 unsigned int ioctl, unsigned long arg)
5534 {
5535 struct kvm_vcpu *vcpu = filp->private_data;
5536 void __user *argp = (void __user *)arg;
5537 int idx;
5538 long r;
5539 u16 rc, rrc;
5540
5541 vcpu_load(vcpu);
5542
5543 switch (ioctl) {
5544 case KVM_S390_STORE_STATUS:
5545 idx = srcu_read_lock(&vcpu->kvm->srcu);
5546 r = kvm_s390_store_status_unloaded(vcpu, arg);
5547 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5548 break;
5549 case KVM_S390_SET_INITIAL_PSW: {
5550 psw_t psw;
5551
5552 r = -EFAULT;
5553 if (copy_from_user(&psw, argp, sizeof(psw)))
5554 break;
5555 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
5556 break;
5557 }
5558 case KVM_S390_CLEAR_RESET:
5559 r = 0;
5560 kvm_arch_vcpu_ioctl_clear_reset(vcpu);
5561 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5562 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5563 UVC_CMD_CPU_RESET_CLEAR, &rc, &rrc);
5564 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET CLEAR VCPU: rc %x rrc %x",
5565 rc, rrc);
5566 }
5567 break;
5568 case KVM_S390_INITIAL_RESET:
5569 r = 0;
5570 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
5571 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5572 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5573 UVC_CMD_CPU_RESET_INITIAL,
5574 &rc, &rrc);
5575 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET INITIAL VCPU: rc %x rrc %x",
5576 rc, rrc);
5577 }
5578 break;
5579 case KVM_S390_NORMAL_RESET:
5580 r = 0;
5581 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
5582 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5583 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5584 UVC_CMD_CPU_RESET, &rc, &rrc);
5585 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET NORMAL VCPU: rc %x rrc %x",
5586 rc, rrc);
5587 }
5588 break;
5589 case KVM_SET_ONE_REG:
5590 case KVM_GET_ONE_REG: {
5591 struct kvm_one_reg reg;
5592 r = -EINVAL;
5593 if (kvm_s390_pv_cpu_is_protected(vcpu))
5594 break;
5595 r = -EFAULT;
5596 if (copy_from_user(®, argp, sizeof(reg)))
5597 break;
5598 if (ioctl == KVM_SET_ONE_REG)
5599 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
5600 else
5601 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
5602 break;
5603 }
5604 #ifdef CONFIG_KVM_S390_UCONTROL
5605 case KVM_S390_UCAS_MAP: {
5606 struct kvm_s390_ucas_mapping ucasmap;
5607
5608 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5609 r = -EFAULT;
5610 break;
5611 }
5612
5613 if (!kvm_is_ucontrol(vcpu->kvm)) {
5614 r = -EINVAL;
5615 break;
5616 }
5617
5618 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
5619 ucasmap.vcpu_addr, ucasmap.length);
5620 break;
5621 }
5622 case KVM_S390_UCAS_UNMAP: {
5623 struct kvm_s390_ucas_mapping ucasmap;
5624
5625 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5626 r = -EFAULT;
5627 break;
5628 }
5629
5630 if (!kvm_is_ucontrol(vcpu->kvm)) {
5631 r = -EINVAL;
5632 break;
5633 }
5634
5635 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
5636 ucasmap.length);
5637 break;
5638 }
5639 #endif
5640 case KVM_S390_VCPU_FAULT: {
5641 r = gmap_fault(vcpu->arch.gmap, arg, 0);
5642 break;
5643 }
5644 case KVM_ENABLE_CAP:
5645 {
5646 struct kvm_enable_cap cap;
5647 r = -EFAULT;
5648 if (copy_from_user(&cap, argp, sizeof(cap)))
5649 break;
5650 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5651 break;
5652 }
5653 case KVM_S390_MEM_OP: {
5654 struct kvm_s390_mem_op mem_op;
5655
5656 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
5657 r = kvm_s390_vcpu_memsida_op(vcpu, &mem_op);
5658 else
5659 r = -EFAULT;
5660 break;
5661 }
5662 case KVM_S390_SET_IRQ_STATE: {
5663 struct kvm_s390_irq_state irq_state;
5664
5665 r = -EFAULT;
5666 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5667 break;
5668 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
5669 irq_state.len == 0 ||
5670 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
5671 r = -EINVAL;
5672 break;
5673 }
5674 /* do not use irq_state.flags, it will break old QEMUs */
5675 r = kvm_s390_set_irq_state(vcpu,
5676 (void __user *) irq_state.buf,
5677 irq_state.len);
5678 break;
5679 }
5680 case KVM_S390_GET_IRQ_STATE: {
5681 struct kvm_s390_irq_state irq_state;
5682
5683 r = -EFAULT;
5684 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5685 break;
5686 if (irq_state.len == 0) {
5687 r = -EINVAL;
5688 break;
5689 }
5690 /* do not use irq_state.flags, it will break old QEMUs */
5691 r = kvm_s390_get_irq_state(vcpu,
5692 (__u8 __user *) irq_state.buf,
5693 irq_state.len);
5694 break;
5695 }
5696 case KVM_S390_PV_CPU_COMMAND: {
5697 struct kvm_pv_cmd cmd;
5698
5699 r = -EINVAL;
5700 if (!is_prot_virt_host())
5701 break;
5702
5703 r = -EFAULT;
5704 if (copy_from_user(&cmd, argp, sizeof(cmd)))
5705 break;
5706
5707 r = -EINVAL;
5708 if (cmd.flags)
5709 break;
5710
5711 /* We only handle this cmd right now */
5712 if (cmd.cmd != KVM_PV_DUMP)
5713 break;
5714
5715 r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd);
5716
5717 /* Always copy over UV rc / rrc data */
5718 if (copy_to_user((__u8 __user *)argp, &cmd.rc,
5719 sizeof(cmd.rc) + sizeof(cmd.rrc)))
5720 r = -EFAULT;
5721 break;
5722 }
5723 default:
5724 r = -ENOTTY;
5725 }
5726
5727 vcpu_put(vcpu);
5728 return r;
5729 }
5730
kvm_arch_vcpu_fault(struct kvm_vcpu * vcpu,struct vm_fault * vmf)5731 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5732 {
5733 #ifdef CONFIG_KVM_S390_UCONTROL
5734 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
5735 && (kvm_is_ucontrol(vcpu->kvm))) {
5736 vmf->page = virt_to_page(vcpu->arch.sie_block);
5737 get_page(vmf->page);
5738 return 0;
5739 }
5740 #endif
5741 return VM_FAULT_SIGBUS;
5742 }
5743
kvm_arch_irqchip_in_kernel(struct kvm * kvm)5744 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
5745 {
5746 return true;
5747 }
5748
5749 /* Section: memory related */
kvm_arch_prepare_memory_region(struct kvm * kvm,const struct kvm_memory_slot * old,struct kvm_memory_slot * new,enum kvm_mr_change change)5750 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5751 const struct kvm_memory_slot *old,
5752 struct kvm_memory_slot *new,
5753 enum kvm_mr_change change)
5754 {
5755 gpa_t size;
5756
5757 if (kvm_is_ucontrol(kvm))
5758 return -EINVAL;
5759
5760 /* When we are protected, we should not change the memory slots */
5761 if (kvm_s390_pv_get_handle(kvm))
5762 return -EINVAL;
5763
5764 if (change != KVM_MR_DELETE && change != KVM_MR_FLAGS_ONLY) {
5765 /*
5766 * A few sanity checks. We can have memory slots which have to be
5767 * located/ended at a segment boundary (1MB). The memory in userland is
5768 * ok to be fragmented into various different vmas. It is okay to mmap()
5769 * and munmap() stuff in this slot after doing this call at any time
5770 */
5771
5772 if (new->userspace_addr & 0xffffful)
5773 return -EINVAL;
5774
5775 size = new->npages * PAGE_SIZE;
5776 if (size & 0xffffful)
5777 return -EINVAL;
5778
5779 if ((new->base_gfn * PAGE_SIZE) + size > kvm->arch.mem_limit)
5780 return -EINVAL;
5781 }
5782
5783 if (!kvm->arch.migration_mode)
5784 return 0;
5785
5786 /*
5787 * Turn off migration mode when:
5788 * - userspace creates a new memslot with dirty logging off,
5789 * - userspace modifies an existing memslot (MOVE or FLAGS_ONLY) and
5790 * dirty logging is turned off.
5791 * Migration mode expects dirty page logging being enabled to store
5792 * its dirty bitmap.
5793 */
5794 if (change != KVM_MR_DELETE &&
5795 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
5796 WARN(kvm_s390_vm_stop_migration(kvm),
5797 "Failed to stop migration mode");
5798
5799 return 0;
5800 }
5801
kvm_arch_commit_memory_region(struct kvm * kvm,struct kvm_memory_slot * old,const struct kvm_memory_slot * new,enum kvm_mr_change change)5802 void kvm_arch_commit_memory_region(struct kvm *kvm,
5803 struct kvm_memory_slot *old,
5804 const struct kvm_memory_slot *new,
5805 enum kvm_mr_change change)
5806 {
5807 int rc = 0;
5808
5809 switch (change) {
5810 case KVM_MR_DELETE:
5811 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5812 old->npages * PAGE_SIZE);
5813 break;
5814 case KVM_MR_MOVE:
5815 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5816 old->npages * PAGE_SIZE);
5817 if (rc)
5818 break;
5819 fallthrough;
5820 case KVM_MR_CREATE:
5821 rc = gmap_map_segment(kvm->arch.gmap, new->userspace_addr,
5822 new->base_gfn * PAGE_SIZE,
5823 new->npages * PAGE_SIZE);
5824 break;
5825 case KVM_MR_FLAGS_ONLY:
5826 break;
5827 default:
5828 WARN(1, "Unknown KVM MR CHANGE: %d\n", change);
5829 }
5830 if (rc)
5831 pr_warn("failed to commit memory region\n");
5832 return;
5833 }
5834
nonhyp_mask(int i)5835 static inline unsigned long nonhyp_mask(int i)
5836 {
5837 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
5838
5839 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
5840 }
5841
kvm_s390_init(void)5842 static int __init kvm_s390_init(void)
5843 {
5844 int i, r;
5845
5846 if (!sclp.has_sief2) {
5847 pr_info("SIE is not available\n");
5848 return -ENODEV;
5849 }
5850
5851 if (nested && hpage) {
5852 pr_info("A KVM host that supports nesting cannot back its KVM guests with huge pages\n");
5853 return -EINVAL;
5854 }
5855
5856 for (i = 0; i < 16; i++)
5857 kvm_s390_fac_base[i] |=
5858 stfle_fac_list[i] & nonhyp_mask(i);
5859
5860 r = __kvm_s390_init();
5861 if (r)
5862 return r;
5863
5864 r = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
5865 if (r) {
5866 __kvm_s390_exit();
5867 return r;
5868 }
5869 return 0;
5870 }
5871
kvm_s390_exit(void)5872 static void __exit kvm_s390_exit(void)
5873 {
5874 kvm_exit();
5875
5876 __kvm_s390_exit();
5877 }
5878
5879 module_init(kvm_s390_init);
5880 module_exit(kvm_s390_exit);
5881
5882 /*
5883 * Enable autoloading of the kvm module.
5884 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
5885 * since x86 takes a different approach.
5886 */
5887 #include <linux/miscdevice.h>
5888 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5889 MODULE_ALIAS("devname:kvm");
5890